<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:googleplay="http://www.google.com/schemas/play-podcasts/1.0"><channel><title><![CDATA[Canonical]]></title><description><![CDATA[Backing founders building for a post AGI future]]></description><link>https://blog.canonical.cc</link><image><url>https://blog.canonical.cc/img/substack.png</url><title>Canonical</title><link>https://blog.canonical.cc</link></image><generator>Substack</generator><lastBuildDate>Sun, 22 Nov 2026 03:53:35 GMT</lastBuildDate><atom:link href="https://blog.canonical.cc/feed" rel="self" type="application/rss+xml"/><copyright><![CDATA[Canonical]]></copyright><language><![CDATA[en]]></language><webMaster><![CDATA[canonicalcc@substack.com]]></webMaster><itunes:owner><itunes:email><![CDATA[canonicalcc@substack.com]]></itunes:email><itunes:name><![CDATA[Canonical]]></itunes:name></itunes:owner><itunes:author><![CDATA[Canonical]]></itunes:author><googleplay:owner><![CDATA[canonicalcc@substack.com]]></googleplay:owner><googleplay:email><![CDATA[canonicalcc@substack.com]]></googleplay:email><googleplay:author><![CDATA[Canonical]]></googleplay:author><itunes:block><![CDATA[Yes]]></itunes:block><item><title><![CDATA[Where the Next 1,000x Runs]]></title><description><![CDATA[The next 1,000x of AI inference won&#8217;t fit in giant data centers. Here&#8217;s where it runs, and where the opportunity is.]]></description><link>https://blog.canonical.cc/p/where-the-next-1000x-runs</link><guid isPermaLink="false">https://blog.canonical.cc/p/where-the-next-1000x-runs</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Wed, 11 Nov 2026 00:53:33 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/f0740b15-5c61-4094-b218-07c5c83f0ce0_1456x762.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><strong>The next 1,000x of AI inference won&#8217;t fit in giant data centers. About 40% of it will run on devices and small sites under 20 MW, and the software that makes those sites work as one network is where the opportunity is.</strong></p><p>Last week&#8217;s midterms were as much a vote on data centers as anything else. Democrats <a href="https://www.nbcnews.com/politics/2026-election/live-blog/midterm-election-results-2026-news-trump-live-updates-rcna605763">took both chambers</a> and <a href="https://www.cnbc.com/2026/11/03/midterm-elections-2026-live-updates.html">flipped the governorships of Texas, Ohio and Georgia</a>. Hinojosa ran on <a href="https://www.spglobal.com/energy/en/news-research/latest-news/electric-power/080426-factbox-energy-affordability-dominates-us-midterm-contests-as-climate-takes-backseat">blocking &#8220;unfair&#8221; data center permits</a>. Voters in 10 states decided <a href="https://news.ballotpedia.org/?p=52078">41 local measures</a> on data centers, 28 of them to limit development.</p><p>The build-out doesn&#8217;t stop. Hyperscaler capex is tracking <a href="https://www.hl.co.uk/news/ai-hyperscalers-are-expected-to-finish-the-year-spending-a-whopping-828-billion-on-capex">above $800B this year</a>, and Amazon says 2027 will be &#8220;meaningfully more.&#8221; What changes is where the fight happens: statehouses, utility commissions and town halls.</p><p>We think the election is a good prompt to ask a bigger question than which gigawatt campus gets its permit. In five years, inference will be 1,000x or more what it is today. Where does all of it run? On phones, laptops and robots? In a rack in an office closet, a 5 MW pod on a solar farm, a 100 MW regional site, a gigawatt campus? What decides which, and how does it spread across countries, cities and homes?</p><p>This post is our attempt at an answer. We built a simple model of cost per token at each kind of site and mapped today&#8217;s workloads onto them. The short version: about a third of 2031&#8217;s tokens still run on giant campuses, a quarter on regional sites, and 41% below the 20 MW line: on devices, in buildings and in small metro sites. Because the total is 1,000x bigger, even the smallest of those slices is many times all the inference running today.</p><p>That&#8217;s where we think the next set of AI infrastructure companies gets built. It may also be the first place crypto has a real job in AI infrastructure. We&#8217;ll be specific about both.</p><h2>Tokens are outgrowing the power behind them</h2><p>Google served 9.7 trillion tokens a month in May 2024, 480 trillion in May 2025, and <a href="https://blog.google/innovation-and-ai/sundar-pichai-io-2026/">3.2 quadrillion in May 2026</a>. That&#8217;s 330x in two years, and still about 7x in the past year alone. Our base case of 1,000x over five years assumes growth slows to about 4x a year. It isn&#8217;t even the bull case.</p><p>But power doesn&#8217;t move that fast. EPRI and Epoch put US AI power capacity at <a href="https://www.spglobal.com/energy/en/news-research/latest-news/electric-power/081325-artificial-intelligence-power-demand-in-us-could-top-50-gw-by-2030-epri">~5 GW today and &gt;50 GW by 2030</a>. The IEA expects US data center electricity use to <a href="https://foxbusiness.com/energy/ai-electricity-demand-surge-also-provide-opportunity-manage-energy">rise ~130% by 2030</a>. Those are big numbers for the grid, but small ones vs. the token curve.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!Cmc0!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!Cmc0!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 424w, https://substackcdn.com/image/fetch/$s_!Cmc0!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 848w, https://substackcdn.com/image/fetch/$s_!Cmc0!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 1272w, https://substackcdn.com/image/fetch/$s_!Cmc0!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!Cmc0!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png" width="1344" height="632" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:632,&quot;width&quot;:1344,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:77978,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/224384822?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!Cmc0!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 424w, https://substackcdn.com/image/fetch/$s_!Cmc0!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 848w, https://substackcdn.com/image/fetch/$s_!Cmc0!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 1272w, https://substackcdn.com/image/fetch/$s_!Cmc0!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6de9c724-4a7b-49b8-9aa6-6664b87f8045_1344x632.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Three things close the gap: </p><ol><li><p>Each watt does more work: AI chip performance per dollar <a href="https://epoch.ai/trends">improves ~49% a year</a>, and the price of a fixed level of capability falls <a href="https://www.a16z.com/llmflation-llm-inference-cost">10x a year</a> by a16z&#8217;s count and <a href="https://epoch.ai/blog/llm-inference-price-trends">a median 50x</a> by Epoch&#8217;s. </p></li><li><p>Small models get good enough fast: open models that fit on one consumer GPU <a href="https://epoch.ai/data-insights/consumer-gpu-model-gap">match the frontier of 6 to 12 months earlier</a>. </p></li><li><p>A lot of the hardware is already bought: billions of phones, laptops and cars carry accelerators that nobody counts as data center capacity.</p></li></ol><p>None of this shrinks demand. Every time the price of a capability drops 10x, new uses show up and total spend rises. Reasoning models went from almost nothing to <a href="https://arxiv.org/abs/2601.10088">over half of</a> <a href="https://arxiv.org/abs/2601.10088">OpenRouter&#8217;s tokens</a> in 2025, and programming went from 11% to &gt;50%. Agents burn tokens the way humans never did. The question isn&#8217;t whether 1,000x shows up. It&#8217;s where it runs.</p><h2>Almost every new rule starts at 20 megawatts</h2><p>The election didn&#8217;t change the demand curve. It changed what it costs to build the biggest sites. Look closely and nearly every new rule targets them.</p><p>The rules already in force (<a href="https://sustainablefutures.linklaters.com/post/102nc7q/new-york-becomes-first-u-s-state-to-impose-a-statewide-data-center-moratorium">New York&#8217;s executive order</a>, <a href="https://www.utilitydive.com/news/texas-law-gives-grid-operator-power-to-disconnect-data-centers-during-crisis/751587">Texas SB 6</a>, <a href="https://www.willkie.com/publications/2026/06/ferc-directs-grid-operators-to-reform-procedures-for-large-loads-connecting-to-the-bulk-transmission">FERC&#8217;s large-load orders</a>, <a href="https://baxtel.com/news/new-jersey-lawmakers-approve-bill-creating-large-load-data-center-tariff">New Jersey&#8217;s tariff</a>) start at between 50-75 MW. The proposals (<a href="https://www.dlapiper.com/insights/publications/2026/07/new-york-state-legislature-passes-first-in-the-nation-data-center-moratorium">New York&#8217;s bill</a>, <a href="https://www.maine.gov/governor/mills/news/governor-mills-announces-decision-ld-307-2026-04-24">Maine&#8217;s vetoed LD 307</a>, the <a href="https://www.softwareseni.com/maine-freezes-the-first-state-level-data-center-moratorium/">Sanders and Ocasio-Cortez moratorium</a>) start at 20 MW.</p><p>The physical queue tells the same story: a big site now waits years for grid capacity, transformers and turbines, then fights the town for its permit. A 12.5 kW node or a 1 MW pod skips that line.</p><p>Small isn&#8217;t unregulated. Urbana, Ohio voted on a <a href="https://www.newsfromthestates.com/node/427338">7.5 MW cap</a>, and Monterey Park&#8217;s citywide ban <a href="https://ballotpedia.org/Monterey_Park,_California,_Measure_NDC,_Prohibit_Data_Centers_Measure_(June_2026)">carves out sub-acre sites</a> that serve a business on the same lot. New Jersey&#8217;s tariff adds up sites under common ownership, which is how a regulator eventually catches a network of small sites. But today, a 5 MW site and a 500 MW campus live under different rulebooks, on different timelines.</p><p>Below 20 MW, you can build faster. But a fast build is worthless if nothing needs to run there. That&#8217;s the real question.</p><h2>Three questions decide where a token runs</h2><p>Every request passes three filters before price matters. Model size decides what can run it, latency decides how far away it can be, and data rules decide where it&#8217;s allowed to go.</p><ol><li><p><strong>Does the model fit?</strong> Weights and KV cache have to sit in fast memory. A phone holds a model of a few billion parameters. A <a href="https://pinggy.io/blog/self_hosting_llms_on_512gb_m5_ultra_mac_studio/">Mac Studio with M5 Ultra</a> holds 512GB at 1.2 TB/s, enough for a 750B-parameter MoE at low precision, and <a href="https://www.engadget.com/ai/you-can-turn-a-cluster-of-macs-into-an-ai-supercomputer-in-macos-tahoe-262-191500778.html">four Mac Studios have run a 1T-parameter model</a>. A <a href="https://www.electronicdesign.com/technologies/embedded/machine-learning/article/55329193/electronic-design-nvidia-jetson-thor-delivers-edge-ai-performance-to-physical-systems">Jetson Thor</a> gives a robot 128GB and 2,070 FP4 TFLOPS on 40 to 130 W. Serving a frontier model with long context to many users takes terabytes of HBM in one NVLink domain. That means NVL72-class racks, and those mean a big site.</p></li><li><p><strong>How fast does the answer have to come back?</strong> A voice agent, a robot arm or a car has a budget of tens of milliseconds, and a cross-country round trip can spend most of it on its own. A coding agent can wait a second. An overnight batch job can wait hours.</p></li><li><p><strong>Where is the data allowed to go?</strong> Patient records, a factory&#8217;s camera feeds, a bank&#8217;s code base, a government&#8217;s citizen data. Rules and contracts pin a lot of work to a building, a company or a border.</p></li></ol><p>Whatever passes all three goes to the cheapest place that can run it. Put together, that gives five tiers.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!7Uze!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!7Uze!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 424w, https://substackcdn.com/image/fetch/$s_!7Uze!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 848w, https://substackcdn.com/image/fetch/$s_!7Uze!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 1272w, https://substackcdn.com/image/fetch/$s_!7Uze!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!7Uze!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png" width="1344" height="806" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/e37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:806,&quot;width&quot;:1344,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:131209,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/224384822?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!7Uze!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 424w, https://substackcdn.com/image/fetch/$s_!7Uze!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 848w, https://substackcdn.com/image/fetch/$s_!7Uze!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 1272w, https://substackcdn.com/image/fetch/$s_!7Uze!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe37cfe61-4448-41ea-b8c3-21708a275072_1344x806.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Frontier training and the largest frontier models stay at the top. Almost everything else could run on more than one tier, which is why cost decides most of the split.</p><h2>What a token costs at each tier</h2><p>For the same job on the same chip, a small site costs 1.3x to 3x what a campus does. That premium is smaller than most people assume, and most of it is idle hardware, not expensive power.</p><p>We built a simple model to check. The same GPU slot sits in every tier: $20K, a four-year life, 2,000 tokens a second at peak. Only the building, the power price, cooling overhead, upkeep and how busy the GPU stays change by tier. Power prices come from EIA: about <a href="https://axis-intelligence.com/us-electricity-statistics/">18&#162;/kWh for homes and 9&#162; for industry</a> this year.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!Yt16!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!Yt16!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 424w, https://substackcdn.com/image/fetch/$s_!Yt16!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 848w, https://substackcdn.com/image/fetch/$s_!Yt16!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 1272w, https://substackcdn.com/image/fetch/$s_!Yt16!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!Yt16!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png" width="1344" height="650" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/d66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:650,&quot;width&quot;:1344,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:84335,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/224384822?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!Yt16!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 424w, https://substackcdn.com/image/fetch/$s_!Yt16!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 848w, https://substackcdn.com/image/fetch/$s_!Yt16!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 1272w, https://substackcdn.com/image/fetch/$s_!Yt16!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd66c3da4-8e1f-4e57-ad8a-741b182d924c_1344x650.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Four things fall out of it.</p><ul><li><p><strong>Hardware is about two thirds of the cost at every tier.</strong> So the expensive thing about a small site is a GPU sitting idle. Moving a metro pod onto stranded solar at 3&#162; only takes it from about 2x to 1.8x a campus.</p></li><li><p><strong>Utilization is the lever.</strong> A home node busy 30% of the time costs about 3x a campus. The same node kept 55% busy by a router that sends it work from other places costs about 1.7x.</p></li><li><p><strong>Devices play a different game.</strong> The owner has already paid for the hardware, so the marginal cost is the electricity: about a cent per million tokens on a phone. Whenever a model fits on the device, it&#8217;s the cheapest place in the system.</p></li><li><p><strong>The absolute numbers are low on purpose.</strong> They leave out networking, staff and margin; an H100 rents for a <a href="https://www.amcompute.com/gpu-cloud-prices/h100">median $3.49 an hour</a> on demand today. The ratios between tiers are the point.</p></li></ul><p>A workload that needs low latency or has to stay in the building will happily pay a 1.3x to 2x premium. A workload with neither will go to the campus. That&#8217;s how we split the 1,000x.</p><h2>How the 1,000x splits in 2031</h2><p>In our base case, 32% of 2031 tokens run on campuses, 27% on regional sites, and 41% below the 20 MW line: 23% on devices, 9% in homes and buildings, 9% in metro pods.</p><p>We got there by sorting tomorrow&#8217;s tokens into seven workloads, running each through the three filters, then letting cost settle the rest. The shares are our judgment, anchored on what OpenRouter and Google report about usage today.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!Dp4c!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!Dp4c!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 424w, https://substackcdn.com/image/fetch/$s_!Dp4c!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 848w, https://substackcdn.com/image/fetch/$s_!Dp4c!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 1272w, https://substackcdn.com/image/fetch/$s_!Dp4c!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!Dp4c!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png" width="1344" height="1168" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/ca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1168,&quot;width&quot;:1344,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:170430,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/224384822?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!Dp4c!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 424w, https://substackcdn.com/image/fetch/$s_!Dp4c!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 848w, https://substackcdn.com/image/fetch/$s_!Dp4c!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 1272w, https://substackcdn.com/image/fetch/$s_!Dp4c!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fca81a754-1a84-4579-b0a6-62ad59073ec9_1344x1168.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Our base case is a judgment call, so we tested what would move it most. Two things matter: how good small models get, and how easy it is to connect small sites. If small models fall further behind and small sites get hard to connect, the share of 2031 tokens running below the 20 MW line drops from 41% to 24% (the centralized case in the chart). If small models catch up faster and new campuses get delayed, that share rises to 54% (the distributed case).</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!nQLz!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!nQLz!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 424w, https://substackcdn.com/image/fetch/$s_!nQLz!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 848w, https://substackcdn.com/image/fetch/$s_!nQLz!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 1272w, https://substackcdn.com/image/fetch/$s_!nQLz!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!nQLz!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png" width="1344" height="632" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:632,&quot;width&quot;:1344,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:78395,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/224384822?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!nQLz!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 424w, https://substackcdn.com/image/fetch/$s_!nQLz!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 848w, https://substackcdn.com/image/fetch/$s_!nQLz!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 1272w, https://substackcdn.com/image/fetch/$s_!nQLz!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94bf44b7-b6b5-43f8-9d49-098f40ca4805_1344x632.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The shares matter less than the size of the pie. If total inference grows 1,000x, the 9% that lands on metro pods is about 90x everything running today. Homes and buildings get about the same. Even in the centralized case, metro pods carry roughly 47x today&#8217;s total. None of these tiers has to win for it to be a very large market. They only have to exist.</p><h2>Small sites only work as a network</h2><p>One small site on its own isn&#8217;t much of a business. A 1 MW pod or a home node goes down, runs out of room, or sits idle when nobody nearby needs it. The cost model says the same thing in numbers: idle hardware is what makes small sites expensive. The value comes from running thousands of them as one network, so each request goes to whichever site is free, close and cheap at that moment. Call it a fleet.</p><p>The first fleets are live, shipping or in the works:</p><ul><li><p><a href="https://www.theinformation.com/articles/nvidias-workaround-data-center-backlash-retrofitting-old-telephone-buildings">Nvidia has pitched Apollo</a> on turning 130+ old telephone buildings, owned by Apollo&#8217;s telecom company Brightspeed, into small AI data centers across up to 20 states. Nvidia puts the spare power at ~1 GW, under 10 MW a building on average.</p></li><li><p>Akamai runs one of the world&#8217;s largest networks for delivering websites and video. It is now putting Nvidia GPUs into that network and selling inference from it across <a href="https://www.crnasia.com/india/news/2026/akamai-takes-ai-inference-to-the-edge-with-nvidia-powered-grid-across-4-400-locations">4,400+ edge locations</a> without building a new site, because the buildings, fiber and customers were already there.</p></li><li><p>SPAN makes smart electrical panels. With homebuilder PulteGroup, it is putting <a href="https://www.businesswire.com/news/home/20260414372626/en/">16-GPU, 12.5 kW nodes</a> into new homes, starting with a 100-home pilot. Each home hosts a slice of a data center, and 8,000 of them add up to 100 MW without ever asking a utility for 100 MW.</p></li><li><p>Nodiac builds modular data centers and, through a deal with PowerBank, is placing <a href="https://www.datacenterdynamics.com/en/news/nodiac-partners-with-powerbank-to-colocate-modular-data-centers-with-solar-and-bess-across-north-america/">1-20 MW pods</a> on solar and battery sites across North America. Those sites already have land, permits and a grid connection, the parts that usually take years.</p></li><li><p>Liquid Energy ships <a href="https://x.com/_LiquidEnergy/status/2118474012464689271">NanoDC</a>, a self-powered container with no grid connection, carrying its own solar panels, batteries and Starlink link. It says a unit can be running anywhere in four weeks. The first units ship this month, with thousands planned from 2027.</p></li></ul><p>Each one gets around the power queue a different way. None of them solves keeping the hardware busy, and that&#8217;s where the network comes in.</p><p>A campus has one owner, one network and one power contract. A fleet has thousands of each, plus uneven power and nobody on site. Three pieces of software are missing:</p><ol><li><p><strong>Routing.</strong> Placing models and sending each request to the right site by latency, power price and what&#8217;s already in cache. This is where utilization comes from, so it&#8217;s where the margin is. Akamai and SPAN are each building their own. Nobody sells it to everyone else.</p></li><li><p><strong>Flexibility.</strong> Emerald AI <a href="https://siliconangle.com/2026/08/25/data-center-power-startup-emerald-ai-raises-150m-at-1-05b-valuation/">raised at a $1.05B valuation</a> selling verified curtailment to big sites, and Silicon Valley Power now <a href="https://thenextweb.com/news/emerald-ai-150m-funding-data-centre-grid-flexibility">trades faster interconnection for it</a>. Nobody does this across 10,000 small sites.</p></li><li><p><strong>Attestation.</strong> Proving a given model ran on a given chip, with the host blind to the data. Confidential computing on Hopper and Blackwell makes it possible. Nobody has packaged it into something an enterprise or a utility will sign off on.</p></li></ol><p>The third is the gate. The enterprise and vision rows in the table above only move off the campus if the buyer can trust a box they don&#8217;t own. Without attestation, distributed inference only gets the workloads nobody cares about.</p><h2>Who owns the edge</h2><p>The further a tier sits from the campus, the more people own it. That matters more than any share in our model, because the number of owners decides how hard a tier is to coordinate and who gets paid.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!_xAu!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!_xAu!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 424w, https://substackcdn.com/image/fetch/$s_!_xAu!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 848w, https://substackcdn.com/image/fetch/$s_!_xAu!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 1272w, https://substackcdn.com/image/fetch/$s_!_xAu!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!_xAu!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png" width="1344" height="880" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:880,&quot;width&quot;:1344,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:105882,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/224384822?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!_xAu!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 424w, https://substackcdn.com/image/fetch/$s_!_xAu!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 848w, https://substackcdn.com/image/fetch/$s_!_xAu!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 1272w, https://substackcdn.com/image/fetch/$s_!_xAu!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F553562ee-13b7-48b2-9f64-bf461d29e292_1344x880.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Rooftop solar is the closest precedent we have. In 2025, small-scale solar made <a href="https://solarquarter.com/2026/02/25/sun-day-campaign-and-eia-data-show-renewables-supplied-nearly-26-of-u-s-electricity-in-2025-led-by-solar-wind-and-battery-storage/">2.06% of US electricity against 6.54% for utility-scale</a>, so ~25% of all solar came from roofs and small sites. Getting there took 15 years, net metering, lease financing and, later, aggregators that bundled homes into VPPs. Our base case puts homes, buildings and metro pods at about 18% of 2031 tokens, in the same range.</p><p>The analogy has limits in both directions. Compute moves over fiber in milliseconds, so a home node can serve a customer three states away; electrons from a roof mostly don&#8217;t leave the street. But a GPU depreciates in 4 years where a solar panel lasts 25, so idle compute is far more expensive than idle panels.</p><p>Geography follows the same logic. Metro pods go where spare distribution capacity, cheap or stranded power and enough nearby users overlap. Premises nodes go where people already are, which is usually where power costs the most. And outside the US, data residency rules push more inference onto regional and metro sites inside each border, especially in countries without hyperscale campuses of their own.</p><h2>Where crypto earns its place</h2><p>Crypto has no job on the campus or regional tiers. A few owners, long contracts and bank financing work fine there. It might have one on the tiers with thousands or millions of owners: homes, buildings, metro pods, and the devices and agents paying into them.</p><p>Bitcoin is the best proof we have that a token can build physical infrastructure. Nobody planned it. A protocol paid strangers for verified work, and they built a global industry of mining companies, data centers and power deals around it, behind an asset that has been worth <a href="https://www.financemagnates.com/cryptocurrency/bitcoin-hits-a-new-high-surpasses-100000-market-cap-exceeds-2-trillion/amp">more than $2T</a>. It has run for 17 years with nobody in charge.</p><p>A fleet of small inference sites has the same three problems Bitcoin solved for hashing. Someone has to fund hardware in thousands of places, often before local demand shows up. Someone has to prove the work was done. And someone has to pay every owner, automatically, in amounts too small for an invoice. Concretely:</p><ul><li><p><strong>Paying many owners.</strong> Millions of nodes and hundreds of operators each get paid per job, often across borders. Agents paying per call for inference are the same problem from the demand side. Stablecoins fit both.</p></li><li><p><strong>Financing hardware.</strong> Hardware is two thirds of cost per token, so the cost of capital is the biggest lever after utilization. On-chain credit against GPUs and pods is an early but real market.</p></li><li><p><strong>Proving what ran.</strong> A buyer routing work to a stranger&#8217;s box needs evidence that the right model ran on the right data, untouched. <a href="https://gensyn.ai/research">Gensyn</a> and <a href="https://www.opengradient.ai/">OpenGradient</a>, both Canonical portfolio companies, work on verifying machine learning work and inference.</p></li></ul><p>The difference is that Bitcoin mining has no customer: the protocol itself is the buyer. Inference needs real demand, and that&#8217;s where today&#8217;s networks fall short. DePIN as a whole made <a href="https://cointelegraph.com/news/depin-revenues-grow-despite-token-losses">~$72M of onchain revenue in 2025</a>, and decentralized compute earns <a href="https://www.yellow.com/research/ai-compute-demand-crypto-gpu-networks-gap-2026">~$200M a year</a>, most of its supply paid for with tokens. Generic GPU marketplaces compete with neoclouds on price and mostly lose.</p><p>The networks that work will pay hosts only for verified work and sell to customers who never touch the token. We&#8217;d drop crypto from this thesis if routing and payments consolidate into a few platforms that settle on cards, if chip vendors&#8217; secure enclaves make attestation good enough without a public ledger, or if homes and buildings stay a rounding error. We&#8217;ll know within a couple of years.</p><h2>What could make us wrong</h2><p>The strongest case against us is that density wins. Industrial power, high utilization and the biggest models all favor the campus, and our own model says a campus is the cheapest place to serve any token that doesn&#8217;t have a reason to be somewhere else. Here&#8217;s what would push the world toward our centralized scenario, where only 24% of tokens run below the 20 MW line:</p><ul><li><p><strong>Small models stall.</strong> If the gap between frontier and small models widens instead of closing, agents and chat escalate to the campus and stay there.</p></li><li><p><strong>Memory runs out.</strong> Devices are limited by DRAM, and DRAM is competing with HBM for the same fabs. Expensive memory caps what fits on a phone or laptop.</p></li><li><p><strong>Hyperscalers build the fleet.</strong> They already run edge networks and modular builds, and could own routing before a startup does.</p></li><li><p><strong>Regulators aggregate.</strong> New Jersey already adds up sites under common ownership. If that spreads, the 20 MW line stops mattering.</p></li><li><p><strong>The unit economics don&#8217;t close.</strong> Retail power, truck rolls and theft across thousands of sites are real costs, and today&#8217;s numbers come from pilots.</p></li><li><p><strong>Demand grows 100x, not 1,000x.</strong> Then campuses have room to spare, and the pressure to put tokens elsewhere fades. The workloads with latency or data constraints still move, but the market is a tenth the size.</p></li><li><p><strong>The election doesn&#8217;t matter.</strong> Capex keeps rising, and builders move to friendlier states, federal land or the Gulf.</p></li></ul><p>The last one is mostly right, and it doesn&#8217;t hurt us. The thesis doesn&#8217;t need a slowdown at the campus, only for big sites to get slower and costlier at the margin while small ones stay easy. Both are already true. Small-model progress and regulatory aggregation are the two we&#8217;ll watch most closely.</p><p>Underneath all of it, the bet is simple: the next 1,000x won&#8217;t fit in giant data centers, and whoever makes small sites work as one network gets paid for it.</p><h2>What we&#8217;d fund</h2><p>We&#8217;re looking for founders building the layers that make thousands of small sites, and billions of devices, work like one cloud:</p><ul><li><p><strong>Routing across tiers.</strong> Software that decides where each request runs, from the phone to the campus, by latency, power price, data rules and what&#8217;s already loaded. Whoever routes sets the utilization, and utilization sets the cost.</p></li><li><p><strong>On-device runtimes.</strong> Compression, scheduling and model tooling that let app developers use the 23% of tokens that will run on hardware their users already own.</p></li><li><p><strong>Proof.</strong> Attestation that a model ran where and how it was claimed, good enough for an enterprise buyer or a utility.</p></li><li><p><strong>Power flexibility.</strong> Software that lets small sites prove to a utility they&#8217;ll cut back on demand, and get connected sooner for it.</p></li><li><p><strong>Modular pods</strong> sold with secured power, remote-operations software and financing, not as bare boxes.</p></li><li><p><strong>Inference hardware</strong> built for sites that can&#8217;t take a liquid-cooled training rack.</p></li><li><p><strong>The physical AI stack.</strong> Inference for robots, vehicles and cameras, where the control loop can&#8217;t wait for a data center.</p></li><li><p><strong>Crypto-native rails</strong> for paying hosts, financing GPUs and settling agent payments, with a product people would buy without the token.</p></li></ul><p>If you&#8217;re building any of this, we&#8217;d love to talk.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Quantum is four businesses. The famous one is last in line.]]></title><description><![CDATA[What we learned going deep on the physics, the hardware and the companies, and where we think the first venture outcomes come from.]]></description><link>https://blog.canonical.cc/p/quantum-is-four-businesses-the-famous</link><guid isPermaLink="false">https://blog.canonical.cc/p/quantum-is-four-businesses-the-famous</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 30 Oct 2026 12:02:59 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/977242eb-349d-4de6-951f-28271f3bb9be_1730x909.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Quantum had a big year. <a href="https://siliconangle.com/2026/06/04/quantum-computer-maker-quantinuum-closes-flat-1-68b-ipo/">Quantinuum raised $1.68B in its IPO</a>. <a href="https://research.google/blog/safeguarding-cryptocurrency-by-disclosing-quantum-vulnerabilities-responsibly/">Google published an estimate</a> that breaking the cryptography behind Bitcoin could take fewer than 500,000 qubits, ~20x fewer than prior estimates. Washington agreed to put <a href="https://finance.yahoo.com/sectors/technology/articles/us-award-2-billion-quantum-102758753.html">$2B into quantum companies in exchange for equity</a>. The easy conclusion is that the quantum computer is about to arrive and everything changes.</p><p>We spent the past week working through the physics, the machines and the companies, and came out with a different view. &#8220;Quantum&#8221; is not one market. <strong>It is four businesses with different customers and different clocks</strong>: computing, sensing, communications, and security. The one everyone talks about, the quantum computer, is the furthest from paying customers. The other three are where we expect the first venture outcomes.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!db_b!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!db_b!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 424w, https://substackcdn.com/image/fetch/$s_!db_b!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 848w, https://substackcdn.com/image/fetch/$s_!db_b!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 1272w, https://substackcdn.com/image/fetch/$s_!db_b!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!db_b!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png" width="1456" height="801" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:801,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!db_b!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 424w, https://substackcdn.com/image/fetch/$s_!db_b!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 848w, https://substackcdn.com/image/fetch/$s_!db_b!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 1272w, https://substackcdn.com/image/fetch/$s_!db_b!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7789696a-e50c-4686-a5e7-1e925011785d_1602x881.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The money tells the same story. <a href="https://quantumconsortium.org/publication/2026-state-of-the-global-quantum-industry-report/">QED-C estimates</a> the whole quantum industry sold $1.9B of products and services in 2025, while raising $4.9B of venture capital and attracting $12.7B of new government commitments. Capital is running well ahead of customers. That is normal for deep tech, but it tells you where to be careful.</p><h2>The one idea you need</h2><p>To see why computing is last in line, you need one idea. It isn&#8217;t &#8220;a qubit is 0 and 1 at the same time.&#8221; <strong>It&#8217;s interference.</strong></p><p>Drop two pebbles in a pond and the ripples overlap. Where two peaks meet, you get a bigger wave. Where a peak meets a trough, the water goes flat.</p><p>Electrons do the same thing. Fire them one at a time at a wall with two slits, and each one lands as a single dot. But after thousands, the screen shows stripes, with bands where almost nothing lands. Opening the second slit actually makes some spots <em>less</em> likely to be hit. Balls can&#8217;t do that. Only waves can.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!CvB8!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!CvB8!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 424w, https://substackcdn.com/image/fetch/$s_!CvB8!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 848w, https://substackcdn.com/image/fetch/$s_!CvB8!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 1272w, https://substackcdn.com/image/fetch/$s_!CvB8!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!CvB8!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png" width="1456" height="585" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:585,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!CvB8!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 424w, https://substackcdn.com/image/fetch/$s_!CvB8!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 848w, https://substackcdn.com/image/fetch/$s_!CvB8!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 1272w, https://substackcdn.com/image/fetch/$s_!CvB8!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F552d8523-083c-4a5d-adc5-3a1b2c994eb9_1894x761.png 1456w" sizes="100vw"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Now put a detector at the slits to see which one each electron goes through. The stripes vanish, and you get two plain piles, like tennis balls. <strong>Measuring the path destroys the interference. </strong>Stripes only form when each electron goes through both slits as a wave and the two halves overlap on the far side. A detector forces the electron to show up at one slit or the other, so there&#8217;s no second half left to overlap with.</p><p>That&#8217;s the whole game. A quantum computer is this experiment built on purpose. It sets up a calculation so the paths to wrong answers cancel out and the paths to the right answer add up, and only then does it measure. It doesn&#8217;t &#8220;try every answer at once.&#8221; It cancels the wrong ones.</p><p>That&#8217;s also why it only helps with problems that have the right structure, like breaking certain encryption or simulating molecules. It doesn&#8217;t help with email, spreadsheets or training today&#8217;s AI models.</p><p>The same physics explains why these machines are so hard to build. Anything that bumps a qubit, whether heat, vibration or a stray magnetic field, acts like that detector at the slits and wipes out the interference.</p><p>Errors also compound fast. A qubit operation that works 99.9% of the time sounds excellent, but chain 1,000 of them together and the chance the whole calculation comes out clean is just 37%. Useful algorithms need millions to billions of operations.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!5hgY!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!5hgY!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 424w, https://substackcdn.com/image/fetch/$s_!5hgY!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 848w, https://substackcdn.com/image/fetch/$s_!5hgY!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 1272w, https://substackcdn.com/image/fetch/$s_!5hgY!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!5hgY!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png" width="1456" height="838" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:838,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!5hgY!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 424w, https://substackcdn.com/image/fetch/$s_!5hgY!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 848w, https://substackcdn.com/image/fetch/$s_!5hgY!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 1272w, https://substackcdn.com/image/fetch/$s_!5hgY!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8bd60bd6-a0b8-45cd-9cde-268c4521dd4b_1892x1089.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The answer is error correction: spread one reliable &#8220;logical&#8221; qubit across many physical ones and keep checking for mistakes without reading the data. This year it started working on a small scale. Quantinuum reported <a href="https://www.quantinuum.com/press-releases/quantinuum-breaks-error-correction-world-record-by-10x-validating-its-leading-path-to-fault-tolerant-quantum-computing-by-2029">99.9995% logical fidelity</a>, and IBM and the University of Chicago <a href="https://newsroom.ibm.com/2026-07-30-ibm-and-the-university-of-chicago-demonstrate-quantum-advantage,-establishing-trusted-quantum-computation-on-logical-circuits">ran 70 logical qubits</a>.</p><p>That progress is real. But the gap is still large. The best machines today have around 100 high-quality physical qubits. Published estimates for breaking RSA-2048 sit <a href="https://arxiv.org/abs/2505.15917">under a million</a>, and useful chemistry needs comparable numbers. The estimates keep falling, which cuts both ways: good for the people building computers, bad for anyone relying on today&#8217;s encryption.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!JN2U!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!JN2U!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 424w, https://substackcdn.com/image/fetch/$s_!JN2U!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 848w, https://substackcdn.com/image/fetch/$s_!JN2U!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 1272w, https://substackcdn.com/image/fetch/$s_!JN2U!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!JN2U!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png" width="1456" height="819" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:819,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!JN2U!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 424w, https://substackcdn.com/image/fetch/$s_!JN2U!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 848w, https://substackcdn.com/image/fetch/$s_!JN2U!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 1272w, https://substackcdn.com/image/fetch/$s_!JN2U!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F19724d2d-5d21-4f44-aff3-5f13810a542b_1930x1085.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><h2>So where does value show up first?</h2><p><strong>The clearest answer is security.</strong> The threat does not need a quantum computer to exist yet. An attacker can record encrypted traffic today and decrypt it later. <a href="https://eprint.iacr.org/2015/1075">Michele Mosca put it as a simple rule</a>: if your data needs to stay secret for ten years, and moving to new encryption takes five, then you needed to start fifteen years before a code-breaking machine arrives. Nobody knows when that is, which is the point.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!YWa8!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!YWa8!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 424w, https://substackcdn.com/image/fetch/$s_!YWa8!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 848w, https://substackcdn.com/image/fetch/$s_!YWa8!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 1272w, https://substackcdn.com/image/fetch/$s_!YWa8!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!YWa8!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png" width="1456" height="831" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:831,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!YWa8!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 424w, https://substackcdn.com/image/fetch/$s_!YWa8!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 848w, https://substackcdn.com/image/fetch/$s_!YWa8!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 1272w, https://substackcdn.com/image/fetch/$s_!YWa8!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388dd9a2-ac57-45cf-a414-f7efcb8a0299_1886x1076.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The new encryption already exists. <a href="https://www.nist.gov/pqc">NIST finalized its post-quantum standards</a> in 2024, and <a href="https://safeheron.com/blog/google-quantum-whitepaper/">Google has set 2029</a> to finish its own migration. The algorithms are free and standardized, so selling the algorithm is not a business.</p><p>The hard part is finding where cryptography lives inside a large organization: certificates, internal APIs, embedded devices, vendor systems, backups. Most companies have no inventory. The startups worth backing own discovery and migration, the system of record for an organization&#8217;s cryptography, not a wrapper around a standard.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!I6bn!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!I6bn!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 424w, https://substackcdn.com/image/fetch/$s_!I6bn!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 848w, https://substackcdn.com/image/fetch/$s_!I6bn!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 1272w, https://substackcdn.com/image/fetch/$s_!I6bn!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!I6bn!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png" width="1456" height="780" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/f3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:780,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!I6bn!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 424w, https://substackcdn.com/image/fetch/$s_!I6bn!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 848w, https://substackcdn.com/image/fetch/$s_!I6bn!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 1272w, https://substackcdn.com/image/fetch/$s_!I6bn!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff3d14d5e-26de-4d1d-8706-2b719be0f82b_1605x860.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Crypto is a special case we care about. Google&#8217;s estimate covers the signatures that secure Bitcoin and Ethereum. Bitcoin <a href="https://blockchaindose.com/post-quantum-crypto-migration/">merged BIP-360</a>, its first quantum-resistant address type, this year, and Ethereum has a post-quantum team. The harder problem is governance, not math: who decides what happens to coins whose keys are already exposed, and how do millions of users migrate?</p><p><strong>Sensing is the second answer</strong>, and it rests on an idea we found useful. The sensitivity that ruins quantum computers is exactly what makes quantum sensors precise. A computer tries to block out the world. A sensor tries to feel it.</p><p>Atomic clocks have been products for decades. The new wave is navigation that works when GPS is jammed, plus magnetometers and gravity sensors, and defense is the first buyer. The test is simple to state and hard to pass: field data, in a package of the right size, weight, power and cost, against the best classical sensor rather than an old one.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!Zu5S!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!Zu5S!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 424w, https://substackcdn.com/image/fetch/$s_!Zu5S!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 848w, https://substackcdn.com/image/fetch/$s_!Zu5S!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 1272w, https://substackcdn.com/image/fetch/$s_!Zu5S!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!Zu5S!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png" width="1456" height="713" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/bc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:713,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!Zu5S!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 424w, https://substackcdn.com/image/fetch/$s_!Zu5S!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 848w, https://substackcdn.com/image/fetch/$s_!Zu5S!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 1272w, https://substackcdn.com/image/fetch/$s_!Zu5S!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbc1fc32d-769e-417a-9ab3-5ab9b7fadcc3_1607x787.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><strong>The third answer is the enabling layer.</strong> Nobody knows which way of building a qubit will win: superconducting circuits, trapped ions, neutral atoms, photons or electron spins. Every one of them needs control electronics, cryogenics, lasers, photonics, calibration software and fast classical processors to decode errors. A company that sells to all of them does not have to pick.</p><p>The risks are a small customer base and the chance that the big hardware makers bring it in-house.</p><p>The part closest to our AI work is AI for quantum: using machine learning to calibrate, control and debug these machines, which can be tested today by how much calibration time it saves.</p><p>We put the longer version on Canonical Labs: an interactive two-slit experiment, an error calculator, a Mosca calculator and the gap between the machines we have and the ones we need. <a href="https://canonical.cc/labs/quantum/">Explore it here</a>. If you are building in post-quantum migration, quantum sensing or the enabling stack, and you want to change how the world works, <strong>we want to meet you.</strong></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[The Cheapest Power Plant Is Off]]></title><description><![CDATA[Backing founders building for a post AGI future]]></description><link>https://blog.canonical.cc/p/the-cheapest-power-plant-is-off</link><guid isPermaLink="false">https://blog.canonical.cc/p/the-cheapest-power-plant-is-off</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Fri, 23 Oct 2026 12:04:08 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/6ade9c8a-a4cd-4f42-9c09-0b0e14b9ce95_1672x941.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Five ideas, papers, and products shaping our investment thinking, and what we think they mean. This series is powered by an AI assistant that helps synthesize recurring themes from our discussions, alongside our own reflections.</em></p><div><hr></div><p><strong><a href="https://www.utilitydive.com/news/google-voltus-vpp-virtual-power-pjm/833717/">Google pays homes and businesses to use less power.</a></strong> The grid serving much of the East Coast is running short of power. Instead of waiting for new plants, Google is paying Voltus to round up batteries, smart thermostats and factory equipment that can cut back when the grid is strained. The deal now covers more than a gigawatt, about one large nuclear reactor&#8217;s worth, through 2040. This matters because data centers can&#8217;t plug in until there&#8217;s power for them, and freeing up existing power is faster than building new plants. We think power you can switch off is becoming something hyperscalers buy, which opens room for startups that find, control and finance those small sources.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/NewsFromGoogle/status/2113269014659027022&quot;,&quot;full_text&quot;:&quot;Today, Google and Voltus announced an expanded agreement to build a 1-gigawatt (GW) virtual power plant across the PJM grid region, which serves 67 million people across 13 states and Washington, D.C. This agreement builds on a 100-megawatt agreement from earlier this year, and&#8230;&quot;,&quot;username&quot;:&quot;NewsFromGoogle&quot;,&quot;name&quot;:&quot;News from Google&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1972717909559353344/eMJ6AJ0W_normal.jpg&quot;,&quot;date&quot;:&quot;2026-10-22T13:59:30.000Z&quot;,&quot;photos&quot;:[],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:7,&quot;retweet_count&quot;:16,&quot;like_count&quot;:270,&quot;impression_count&quot;:11899,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.anthropic.com/research/designing-efficient-quantum-circuits">Claude designed up to 10x better quantum circuits.</a></strong> Quantum circuits are the step-by-step programs a quantum computer runs. The fewer qubits and steps they need, the sooner real hardware can run them. Anthropic gave Claude eight standard circuits and a program that scored each attempt. With no human help, Claude beat the best published design on seven of the eight. This matters because AI gets better fastest wherever it can check its own work instantly and cheaply. We think the bigger story is crypto: one of these circuits is a key step in breaking the encryption behind crypto wallets, which makes post-quantum upgrades more urgent.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/AnthropicAI/status/2111152183240511640&quot;,&quot;full_text&quot;:&quot;New from our STEM Fellows Program: Designing more efficient quantum circuits with Claude.\n\nGiven plain-language instructions, Claude Mythos 5 designed quantum circuits that improve on the best published ones, sometimes by an order of magnitude or more.\n\nMore efficient circuits&#8230;&quot;,&quot;username&quot;:&quot;AnthropicAI&quot;,&quot;name&quot;:&quot;Anthropic&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1798110641414443008/XP8gyBaY_normal.jpg&quot;,&quot;date&quot;:&quot;2026-10-16T17:47:58.000Z&quot;,&quot;photos&quot;:[],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:484,&quot;retweet_count&quot;:615,&quot;like_count&quot;:5497,&quot;impression_count&quot;:913088,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://odyssey.systems/meet-odyssey-3">Odyssey powered a self-driving car with only 20 hours of driving data.</a></strong> Waymo and Tesla needed years and millions of miles of data. Odyssey&#8217;s model first learned how the world works by watching video and games, so it needed only a little real driving to handle roads in India. The same model also became the base for robot arms and a humanoid. This matters because if one model can learn many machines from little data, robotics stops being a costly data race for each one. We think world models become the foundation of robotics, and startups win by building on them instead of collecting data from scratch.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/odysseyml/status/2111589148146749616&quot;,&quot;full_text&quot;:&quot;We placed the fruit fly brain inside Odyssey-3, a frontier world model.\n\nThe lil&#8217; guy is now flying around totally imagined, simulated worlds all by itself &#129782; &quot;,&quot;username&quot;:&quot;odysseyml&quot;,&quot;name&quot;:&quot;Odyssey&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2071642815944359936/IebkUQ-C_normal.jpg&quot;,&quot;date&quot;:&quot;2026-10-17T22:44:19.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!sq2_!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2111589052428500992.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/lvKrGoTtKr&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:674,&quot;retweet_count&quot;:1786,&quot;like_count&quot;:20442,&quot;impression_count&quot;:6231894,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2111589052428500992/vid/avc1/1280x720/SIQRinxd5ZChXYlN.mp4&quot;,&quot;video_preview_media_key&quot;:&quot;13_2111589052428500992&quot;,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://arxiv.org/abs/2610.29677">Robots are learning from their mistakes, not yet learning how to learn.</a></strong> It&#8217;s the difference between a student reviewing wrong answers and a student finding a better way to study. A new 93-page survey finds robots now do the first, but no study yet shows the second. Software can attempt it because it tests every change instantly, while robots pay for every try in time, money and sometimes broken hardware. This matters because &#8220;self-improving robots&#8221; is becoming a pitch that usually means only the first. We think the teams worth backing can show, and measure, their robots getting better with every round of deployment.</p><div><hr></div><p><strong><a href="https://already-clean.goodfire.com/">AI models may be much easier to understand than we thought.</a></strong> The common view is that each &#8220;neuron&#8221; inside a model mixes many unrelated concepts, which makes models hard to read. Goodfire&#8217;s Nick Cammarata argues that is mostly a measurement problem: in an early check of one open model, about 90% of neurons had one clear job. This matters because a model you can read is a model you can audit and fix. We think that helps open-weight models most, since anyone can inspect them, which gives them an edge wherever regulators want to know how a model works.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/MaxForAI/status/2110923965107704110&quot;,&quot;full_text&quot;:&quot;&#26377;&#28857;&#38663;&#25788;&#12290;&#12290;&#12290;&#25105;&#20204;&#21487;&#33021;&#30495;&#30340;&#24555;&#35201;&#25630;&#28165;&#26970;&#22823;&#27169;&#22411;&#20869;&#37096;&#21040;&#24213;&#26159;&#24590;&#20040;&#36816;&#20316;&#30340;&#20102;&#12290;\n\nOpenAI&#26089;&#26399;&#26426;&#26800;&#21487;&#35299;&#37322;&#24615;&#30740;&#31350;&#30340;&#21442;&#19982;&#32773;&#65292;&#26152;&#22825;&#31361;&#28982;&#25552;&#20986;&#20102;&#19968;&#20010;&#24456;&#22823;&#32966;&#30340;&#32467;&#35770;&#65306;\n\n&#25105;&#20204;&#21487;&#33021;&#19968;&#30452;&#35823;&#35299;&#20102;&#31070;&#32463;&#32593;&#32476;&#12290;\n\n&#26152;&#22825;&#65292; <span class=\&quot;tweet-fake-link\&quot;>@nickcammarata</span>  &#21457;&#20102;&#19968;&#31687;&#24456;&#26377;&#24847;&#24605;&#30340;&#25991;&#31456;&#65292;&#21483;&#12298;Neural Networks Are Already Clean&#12299;&#12290;&#8230;&quot;,&quot;username&quot;:&quot;MaxForAI&quot;,&quot;name&quot;:&quot;Max For AI&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1885611457376903171/zWWnahYp_normal.jpg&quot;,&quot;date&quot;:&quot;2026-10-16T02:41:07.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HUuCHvmbMAAUG1B.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/T6dYgGeFs2&quot;}],&quot;quoted_tweet&quot;:{&quot;full_text&quot;:&quot;interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came&quot;,&quot;username&quot;:&quot;nickcammarata&quot;,&quot;name&quot;:&quot;Nick&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1753264923365523456/mUCvwn7v_normal.jpg&quot;,&quot;photos&quot;:[]},&quot;reply_count&quot;:58,&quot;retweet_count&quot;:445,&quot;like_count&quot;:2971,&quot;impression_count&quot;:255714,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><em>We&#8217;ll share another edition next week.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Brains Are Cheap, Hands Are Not]]></title><description><![CDATA[AI models fail at household chores, a cheap way to train robots, why cost is the real bottleneck in factories, the memory problem in AI chips, and AI shopping agents that misread what people]]></description><link>https://blog.canonical.cc/p/brains-are-cheap-hands-are-not</link><guid isPermaLink="false">https://blog.canonical.cc/p/brains-are-cheap-hands-are-not</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 09 Oct 2026 12:15:13 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/8a4e1fcf-4827-42d8-a752-3488875c5434_1672x941.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Five ideas, papers, and products shaping our investment thinking, and what we think they mean. This series is powered by an AI assistant that helps synthesize recurring themes from our discussions, alongside our own reflections.</em></p><div><hr></div><p><strong><a href="https://arxiv.org/abs/2610.00854">Today&#8217;s best AI models can control a robot arm, but they all failed at household chores.</a></strong> Researchers tested seven leading AI models at controlling robot arms across about 1,000 simulated tasks. The best model, GPT-6 Astra, finished 42% pick-and-place tasks. But on household chores, where the robot has to move several objects around a room in a sequence, every model failed every time. We think LLMs that decide what a robot should do are improving fast and will soon be cheap. The value is in getting the robot to actually carry it out reliably in the real world, i.e. control, sensing, and data from real deployments.</p><div><hr></div><p><strong><a href="https://arxiv.org/abs/2609.38982">A new method taught an LLM-driven robot to fold towels and handle plates with 10 minutes of real-world practice.</a></strong> Learning by trial and error on real robots is slow and expensive. So researchers at Amazon FAR, UT Austin and CMU had a coding agent practice in a rough simulation first, then gave it five tries per task on a real robot. After each failure, the agent tested fixes in the simulator and kept only the ones that worked. It succeeded on 26 of 30 real attempts, while the best alternative managed 3. We think this is a real unlock: teaching robots new tasks normally takes hours of robot time and lots of real data, and this needed a rough sim and a few real tries.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/JiahengHu1/status/2107485908673106241&quot;,&quot;full_text&quot;:&quot;We&#8217;ve all seen the cool &#8220;Astra for robotics&#8221; demos. But behind the scenes? Slow rollouts, hand-engineered APIs, and/or simulated behaviors that struggle on real robots.\n\nHow do we make coding agent actually work in the physical world? &#129470; Introducing SimEX: Simulation-Integrated &#8230;&quot;,&quot;username&quot;:&quot;JiahengHu1&quot;,&quot;name&quot;:&quot;Jiaheng Hu&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2091388160307601408/uey3hzrv_normal.jpg&quot;,&quot;date&quot;:&quot;2026-10-06T14:59:30.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!3HSb!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2107485879019405313.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/DkjHs392kl&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:11,&quot;retweet_count&quot;:60,&quot;like_count&quot;:290,&quot;impression_count&quot;:28993,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2107485879019405313/vid/avc1/1280x720/NEXEsHXpdmgoyQt_.mp4?tag=16&quot;,&quot;video_preview_media_key&quot;:&quot;13_2107485879019405313&quot;,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.anthropic.com/research/what-work-can-robots-do">Where robots already work reliably, in factories and warehouses, the problem isn&#8217;t capability. It&#8217;s cost.</a></strong> Anthropic&#8217;s economists rated 7,594 physical job tasks on whether a robot can do them today and whether it would cost less than a human. Robots can handle about three-quarters of them, almost entirely in controlled settings like factory lines and warehouses; in messy environments like homes, that drops to about 1%. Even where robots can do the work, they&#8217;re cheaper than a person for only 0.3% of tasks. We think the winners will be the companies that make the unit economics work, e.g. robots-as-a-service priced per task instead of hardware sold up front.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/rclegateyang/status/2105329450229608841&quot;,&quot;full_text&quot;:&quot;New report with <span class=\&quot;tweet-fake-link\&quot;>@AnthropicAI</span>: What work can robots do today?\n\nWe find that robots extend AI exposure far beyond digital work. Over 80% of working hours in the US are exposed to robots or LLMs, but robots are cost-competitive for just 0.3%.\n\nIf robot price declines follow past &#8230;&quot;,&quot;username&quot;:&quot;rclegateyang&quot;,&quot;name&quot;:&quot;Russell Legate-Yang&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1780697970184265728/dMx6HCwW_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-30T16:10:30.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HTeh-j1aYAEKQsz.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/XWsguZw5Z3&quot;},{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HTeh-jzbYAAjjB0.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/XWsguZw5Z3&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:18,&quot;retweet_count&quot;:46,&quot;like_count&quot;:300,&quot;impression_count&quot;:38745,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.youtube.com/watch?v=OpeCP4wCxkA">The bottleneck for AI inference chips is memory bandwidth, not compute.</a></strong> On No Priors, Fractile&#8217;s CEO argued that running large models is limited by how fast a chip can move data out of memory. Over 20 years, compute grew about a millionfold while memory bandwidth grew only ~40x. Fractile claims its custom DRAM design delivers ~25x the bandwidth of today&#8217;s HBM-based chips. We think memory architecture will increasingly decide the cost of inference. We&#8217;re looking for memory-first chip startups and for tools that shorten chip design cycles.</p><div id="youtube2-OpeCP4wCxkA" class="youtube-wrap" data-attrs="{&quot;videoId&quot;:&quot;OpeCP4wCxkA&quot;,&quot;startTime&quot;:null,&quot;endTime&quot;:null}" data-component-name="Youtube2ToDOM"><div class="youtube-inner"><iframe src="https://www.youtube-nocookie.com/embed/OpeCP4wCxkA?rel=0&amp;autoplay=0&amp;showinfo=0&amp;enablejsapi=0" frameborder="0" loading="lazy" gesture="media" allow="autoplay; fullscreen" allowautoplay="true" allowfullscreen="true" width="728" height="409"></iframe></div></div><div><hr></div><p><strong><a href="https://www.anthropic.com/research/project-swap">Anthropic let Claude agents trade on behalf of 201 employees, and their biggest problem was understanding what people wanted.</a></strong> In Project Swap, each agent interviewed its owner about their reading tastes, then traded used books with the other agents. On average, people ended up with about their fifth-favorite book out of ten. About 85% of the shortfall came from agents misjudging preferences, not from bad trading. We think agentic commerce is bottlenecked on understanding the user, not on payment rails. Stablecoins can handle settlement; the edge goes to whoever models customer intent best.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/AnthropicAI/status/2047728360818696302&quot;,&quot;full_text&quot;:&quot;New Anthropic research: Project Deal.\n\nWe created a marketplace for employees in our San Francisco office, with one big twist. We tasked Claude with buying, selling and negotiating on our colleagues&#8217; behalf. &quot;,&quot;username&quot;:&quot;AnthropicAI&quot;,&quot;name&quot;:&quot;Anthropic&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1798110641414443008/XP8gyBaY_normal.jpg&quot;,&quot;date&quot;:&quot;2026-04-24T17:24:20.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!F8zD!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2047716403000606720.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/H2f6cLDlAW&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:462,&quot;retweet_count&quot;:704,&quot;like_count&quot;:7484,&quot;impression_count&quot;:2915651,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2047716403000606720/vid/avc1/720x900/upk-uNdEcmY6Yb_l.mp4&quot;,&quot;video_preview_media_key&quot;:&quot;13_2047716403000606720&quot;,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><em>We&#8217;ll share another edition next week.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Power Is the Rent]]></title><description><![CDATA[AI power leases turn into bonds, regulators slow compute futures, agents start designing chips, robots stall on cost, and stablecoins become B2B rails.]]></description><link>https://blog.canonical.cc/p/power-is-the-rent</link><guid isPermaLink="false">https://blog.canonical.cc/p/power-is-the-rent</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Fri, 02 Oct 2026 12:16:19 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/23157fa4-03dd-47bd-8514-04138abbe8ca_1672x941.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Five ideas, papers, and products shaping our investment thinking, and what we think they mean. This series is powered by an AI assistant that helps synthesize recurring themes from our discussions, alongside our own reflections.</em></p><div><hr></div><p><strong><a href="https://www.sec.gov/Archives/edgar/data/0000827876/000119312526402944/clsk-ex99_1.htm">AI power is being underwritten like infrastructure.</a></strong> CleanSpark closed $2.276B of senior secured notes, following its <a href="https://www.sec.gov/Archives/edgar/data/0000827876/000119312526302448/clsk-ex99_1.htm">20-year, 175 MW lease</a> with an unnamed investment-grade tenant worth $6.6B. TeraWulf signed a similar <a href="https://investors.terawulf.com/news-events/press-releases/detail/142/">20-year, 401 MW lease with Anthropic</a> worth $19B. This matters because the credit belongs to the tenant while the scarce asset is the energized site, which is why former bitcoin miners now borrow like utilities. We think the opportunity sits between power and compute: project-level financing, and capacity for tenants below investment grade that still need megawatts.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/Pep_Invest/status/2077090755550872003&quot;,&quot;full_text&quot;:&quot;$CLSK just secured a massive 20-year contract for its Sandersville, Georgia site.\n\nIt guarantees $6.6 billion in revenue (up to $11.6 billion with extensions).\n\n$CLSK is renting out its infrastructure for AI and HPC\n\nCrucially, the buyer also locked in exclusive rights to $CLSK&#8230;&quot;,&quot;username&quot;:&quot;Pep_Invest&quot;,&quot;name&quot;:&quot;Pep Invest&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2094791851131740160/lqBL8hC2_normal.jpg&quot;,&quot;date&quot;:&quot;2026-07-14T18:00:01.000Z&quot;,&quot;photos&quot;:[],&quot;quoted_tweet&quot;:{&quot;full_text&quot;:&quot;$CLSK CleanSpark just locked in a 20-year, $6.6B lease (175 MW) with a high-grade tech giant, plus exclusivity for another 885 MW in Texas.\n\nAfter the massive run-up in semis, I&#8217;m shifting focus back to neoclouds - demand for critical power is so high that every megawatt gets&quot;,&quot;username&quot;:&quot;FinnStockinger&quot;,&quot;name&quot;:&quot;Finn Stockinger&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2013188852278517760/rPZTZ1cG_normal.jpg&quot;},&quot;reply_count&quot;:2,&quot;retweet_count&quot;:1,&quot;like_count&quot;:9,&quot;impression_count&quot;:3492,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.cftc.gov/filings/documents/2026/orgdcmnymexcompcontr260921.pdf">Compute futures hit a regulatory speed bump.</a></strong> The CFTC extended its review of CME&#8217;s proposed H100 and B200 rental index futures to November 9, citing novel or complex issues. A separate <a href="https://www.cftc.gov/PressRoom/PressReleases/9286-26">request for comment</a> asks about cash-market liquidity, manipulation and customer protection. This matters because compute can&#8217;t be hedged or financed at scale until there&#8217;s a price people trust. We think the opening is index integrity: neutral, transaction-based benchmarks and verifiable settlement underneath any forward curve, GPU-backed loan or onchain compute market.</p><div><hr></div><p><strong><a href="https://news.synopsys.com/2026-09-30-OpenAI-and-Synopsys-Announce-GPT-Synopsys-Frontier-Intelligence-to-Revolutionize-Chip-Design">OpenAI is building a chip-design model with Synopsys.</a></strong> GPT-Synopsys will operate Synopsys&#8217;s EDA tools directly under a multi-year revenue-sharing deal, without training on customer designs. The same day, Synopsys signed a <a href="https://news.synopsys.com/2026-09-30-Synopsys-and-Amazon-Announce-Strategic,-Multi-year-IP-Agreement-for-Custom-Silicon-Collaboration-Also-Extends-to-Cloud-and-AI-Powered-Engineering">$1B-plus IP deal with Amazon</a> that also covers its AI tools. This matters because a frontier lab now sees chip design as worth its own model, and chose to enter through the incumbent. We think a model trained to drive one vendor&#8217;s tools learns the tools, not the judgment of the designers using them. That leaves room for an independent layer built on real design decisions that works across every vendor&#8217;s flow.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/Synopsys/status/2105365065058902447&quot;,&quot;full_text&quot;:&quot;&#128680; Synopsys and <span class=\&quot;tweet-fake-link\&quot;>@OpenAI</span> are partnering to create GPT-Synopsys, a specialized frontier model for AI-native chip design, built using OpenAI's frontier intelligence with Synopsys' trusted EDA technology and deep domain expertise. Learn more: <a class=\&quot;tweet-url\&quot; href=\&quot;https://bit.ly/46Y1SBl\&quot;>bit.ly/46Y1SBl</a> &quot;,&quot;username&quot;:&quot;Synopsys&quot;,&quot;name&quot;:&quot;Synopsys&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/959597157601878016/bfge93-W_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-30T18:32:01.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!JR0R!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2105364391919235072.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/k1MTBLvY5n&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:16,&quot;retweet_count&quot;:32,&quot;like_count&quot;:300,&quot;impression_count&quot;:124011,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2105364391919235072/vid/avc1/1280x720/XKq2jzTDHD6zODRu.mp4&quot;,&quot;video_preview_media_key&quot;:&quot;13_2105364391919235072&quot;,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.anthropic.com/research/what-work-can-robots-do">Robots can do the work, just not cheaply enough.</a></strong> Anthropic estimates robots could perform 74% of physical job tasks, but only 0.3% are cost-competitive with human labor today. Training is getting cheaper: <a href="https://arxiv.org/abs/2609.39870">Magic-W0</a>, a new world-action model, learns from human video, real-robot and simulation data and adapts to real tasks with little fine-tuning. This matters because physical AI&#8217;s constraint has moved from capability to unit economics. We think the nearer-term winners are cost-curve bets: cheaper hardware, structured environments, and the financing and integration layers that make deployment pencil out.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/PeterMcCrory/status/2105366155917656502&quot;,&quot;full_text&quot;:&quot;What work can robots do today? And what might that imply for the structure of the economy in the years ahead?\n\nNew research from <span class=\&quot;tweet-fake-link\&quot;>@rclegateyang</span> and Maxim Massenkoff out today exploring these questions. &quot;,&quot;username&quot;:&quot;PeterMcCrory&quot;,&quot;name&quot;:&quot;Peter McCrory&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2011984403191500800/KLJy_IKI_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-30T18:36:22.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HTe-R4BbUAAZFkg.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/mMNcqGP77t&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:4,&quot;retweet_count&quot;:17,&quot;like_count&quot;:60,&quot;impression_count&quot;:6327,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.allium.so/reports/state-of-stablecoins-and-payments-september-2026">Stablecoins are becoming B2B payment rails.</a></strong> Allium counts $401-527B in stablecoin payments through August, up 42-63% year over year, with business-to-business the largest lane at $137-153B. Supply grew only 6%. This matters because volume, not supply, is now the story: stablecoins are settling real corporate flows. We think the next opportunity is credit on top of the rails: short-duration working capital for payment companies and their suppliers, financed onchain, where repayment data already sits in the transaction history.</p><div><hr></div><p><em>We&#8217;ll share another edition next week.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Nobody Wants to Be Middleware]]></title><description><![CDATA[Amazon blocks Meta's shopping agent, BlackRock bets on machine money, and 950 Claude agents find a new enzyme system in 21 hours.]]></description><link>https://blog.canonical.cc/p/nobody-wants-to-be-middleware</link><guid isPermaLink="false">https://blog.canonical.cc/p/nobody-wants-to-be-middleware</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Fri, 25 Sep 2026 17:50:37 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/fb345de0-efa9-462e-901a-390aa827661d_1672x941.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Five ideas, papers, and products shaping our investment thinking, and what we think they mean.</em> <em>This series is powered by an AI assistant that helps synthesize recurring themes from our discussions, alongside our own reflections.</em></p><div><hr></div><p><strong>Amazon blocked Meta&#8217;s shopping agent, and its rivals rushed to plug in.</strong> Meta launched <a href="https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/">Muse</a> on September 8, a personal agent that shops, books, and pays from its own cloud computer, and it went straight to the top of the App Store. <a href="https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/">Amazon blocked it</a>, but Walmart, Shopify, PayPal, and Expedia integrated. This matters because an agent never sees a sponsored listing, and Amazon made more than $68 billion from ads last year. We think this is the playbook in every category: incumbents block agents, challengers embrace them to take share, and incumbents come around later on worse terms. Most exposed are the businesses that monetize inertia, like subscriptions nobody cancels and insurance nobody reprices.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/jwegener/status/2101900045528535095&quot;,&quot;full_text&quot;:&quot;Muse vs Amazon bot detection. Uh oh &quot;,&quot;username&quot;:&quot;jwegener&quot;,&quot;name&quot;:&quot;Jonathan Wegener&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/378800000646555393/98dcd09e535f86adb10e72feea80eece_normal.jpeg&quot;,&quot;date&quot;:&quot;2026-09-21T05:03:16.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HSty87kakAAOgSP.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/rJcZ7ufqMq&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:10,&quot;retweet_count&quot;:2,&quot;like_count&quot;:123,&quot;impression_count&quot;:26650,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong>BlackRock says agents will pay in stablecoins. The biggest agent launch of the year pays with cards.</strong> BlackRock published <a href="https://www.disruptionbanking.com/2026/09/25/blackrock-paper-says-ai-agents-could-add-demand-for-stablecoins/">The Machine-Native Economy</a> this week, arguing that agents need programmable, always-on money and naming stablecoins, Ethereum, and Circle&#8217;s Arc as the likely rails. Meanwhile, Muse checks out through <a href="https://stripe.com/newsroom/news/stripe-helps-meta-muse-shop-with-link">Stripe&#8217;s Link wallet</a>, using a saved card or a single-use virtual one. This matters because consumer agents inherit the rails of the person they act for, and that person already has a card on file. We think stablecoins win where there&#8217;s no human account behind the transaction at all: agents paying APIs, other agents, and compute per call, across borders and in amounts too small for card economics.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/BlackRock/status/2102409739175141458&quot;,&quot;full_text&quot;:&quot;Our latest research paper explores the growing connection between AI and digital assets and explains why broad AI adoption may drive new demand, utility and applications across the digital asset economy. <a class=\&quot;tweet-url\&quot; href=\&quot;https://www.blackrock.com/us/individual/literature/whitepaper/the-machine-native-economy.pdf\&quot;>blackrock.com/us/individual/&#8230;</a> &quot;,&quot;username&quot;:&quot;BlackRock&quot;,&quot;name&quot;:&quot;BlackRock&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1675852182183190528/EubQyAvV_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-22T14:48:37.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HS1CHTXWIAArMC2.png&quot;,&quot;link_url&quot;:&quot;https://t.co/GMxQqTVmUN&quot;},{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HS1CHTfXIAARatK.png&quot;,&quot;link_url&quot;:&quot;https://t.co/GMxQqTVmUN&quot;},{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HS1CHTcXUAAvEfL.png&quot;,&quot;link_url&quot;:&quot;https://t.co/GMxQqTVmUN&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:744,&quot;retweet_count&quot;:1925,&quot;like_count&quot;:8156,&quot;impression_count&quot;:3792108,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong>950 Claude agents found a new CRISPR-like enzyme system in 21 hours.</strong> Anthropic&#8217;s new life sciences lab <a href="https://www.anthropic.com/news/claude-discovers-novel-enzyme-system">pointed Claude at a DNA database</a> with a single prompt. Roughly 950 agents sorted more than 200,000 enzymes into 20 written reports, and one spotted a repeating DNA pattern beside an unusual enzyme that nobody had described, reminiscent of CRISPR. This matters because genome mining that used to take an expert weeks to months now takes a day of compute, and hypotheses are becoming abundant. We think the bottleneck in science is moving from generating ideas to verifying them, and the durable value sits with whoever owns the validation loop: wet labs, automated experimentation, and the taste to decide what&#8217;s worth testing.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/AnthropicAI/status/2102824961538920822&quot;,&quot;full_text&quot;:&quot;This is the first result from our new molecular biology lab, where a team of Anthropic biologists is using Claude to explore and accelerate fundamental biology research. There, Claude works through data and literature to generate hypotheses and candidate biological systems to &#8230;&quot;,&quot;username&quot;:&quot;AnthropicAI&quot;,&quot;name&quot;:&quot;Anthropic&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1798110641414443008/XP8gyBaY_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-23T18:18:34.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!OkA_!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2102822016328028160.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/2rlsYxGowp&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:167,&quot;retweet_count&quot;:280,&quot;like_count&quot;:4801,&quot;impression_count&quot;:979216,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2102822016328028160/vid/avc1/1188x720/E-pVb0T4bKE6tBzO.mp4&quot;,&quot;video_preview_media_key&quot;:&quot;13_2102822016328028160&quot;,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong>Robots are generating data nobody can store.</strong> Felicis argues <a href="https://www.felicis.com/blog/the-physical-world-has-no-data-stack">the physical world has no data stack</a>. A single autonomous vehicle can produce terabytes a day of synchronized video, LiDAR, and telemetry, and most of it gets sampled down or deleted, because today&#8217;s tools were built for rows and columns. Foxglove, whose open MCAP format is the default logger in ROS 2, is the early leader. This matters because every fleet that scales becomes a data infrastructure customer, and every robot foundation model needs that data to improve. We think the pre-seed openings are at the edges: deciding on the robot what&#8217;s worth keeping, and making physical-world data portable enough to share and sell, because data that can&#8217;t move can&#8217;t compound.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!9vcn!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!9vcn!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 424w, https://substackcdn.com/image/fetch/$s_!9vcn!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 848w, https://substackcdn.com/image/fetch/$s_!9vcn!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 1272w, https://substackcdn.com/image/fetch/$s_!9vcn!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!9vcn!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png" width="1200" height="900" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:900,&quot;width&quot;:1200,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:&quot;Every new shape of data has produced its own infrastructure company. &quot;,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="Every new shape of data has produced its own infrastructure company. " title="Every new shape of data has produced its own infrastructure company. " srcset="https://substackcdn.com/image/fetch/$s_!9vcn!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 424w, https://substackcdn.com/image/fetch/$s_!9vcn!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 848w, https://substackcdn.com/image/fetch/$s_!9vcn!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 1272w, https://substackcdn.com/image/fetch/$s_!9vcn!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9a00ee60-58f1-489f-91f7-925ca7d5516e_1200x900.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-GK1x3M" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-GK1x3M" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><div><hr></div><p><strong>DeFi built the right tools for the wrong assets.</strong> Multicoin&#8217;s new essay on <a href="https://multicoin.capital/2026/09/24/defi-2-0/">DeFi 2.0</a> argues that order books, fixed-rate lending, and portfolio margin have existed onchain for years, but were built for volatile, perpetual cryptoassets that didn&#8217;t need them. Treasuries, equities, and credit have maturities, cash flows, and identifiable borrowers, which is exactly what those primitives were designed for. This matters because the RWA conversation has been stuck on tokenization, when the value is in what happens after an asset arrives. We think the best early bets are primitives whose market grows with the asset base, and fixed-term, underwritten credit for real borrowers is the one we care about most, because it&#8217;s what hardware and robotics credit needs to move onchain.</p><div><hr></div><p><em>We&#8217;ll share another edition next week.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Robots Don't Need Robots]]></title><description><![CDATA[One model learned from a glove, another from other robots, a third from video. None of them needed a teleop fleet.]]></description><link>https://blog.canonical.cc/p/robots-dont-need-robots</link><guid isPermaLink="false">https://blog.canonical.cc/p/robots-dont-need-robots</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Fri, 18 Sep 2026 12:15:15 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!5iDQ!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2099897328866115585.jpg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Five ideas, papers, and products shaping our investment thinking, and what we think they mean.</em> <em>This series is powered by an AI assistant that helps synthesize recurring themes from our discussions, alongside our own reflections.</em></p><div><hr></div><p><strong>Three labs bet in the same week that teleoperation data is not the moat.</strong> <a href="https://odyssey.systems/introducing-odyssey-3">Odyssey-3</a> controls arms, humanoids, cars, drones and video games from one pretrained world model with tens of hours of task-specific data, and its sim-trained driving policies traveled about 77% as far between safety interventions as policies trained on real footage. <a href="https://www.rewardai.com/blog/OM-1/">Reward AI&#8217;s OM-1</a> skips robot data entirely, learning from people wearing a capture glove and transferring zero-shot from tabletop arms to humanoids. <a href="https://zeno-3d-web.vercel.app/research/zeno-1-collaborative-intelligence">Zeno-1</a> trains robots on each other, using four hours of closed-loop partner interaction to run decentralized multi-robot collaboration at 30Hz. This matters because much of the industry has spent three years assuming proprietary teleop data was a defensible asset, and three different technical routes just suggested otherwise. We think the moat moves down to deployment, meaning customer contracts, fleets in the field, and the service relationships that produce revenue rather than demos.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/odysseyml/status/2099900067356586276&quot;,&quot;full_text&quot;:&quot;Today we&#8217;re unveiling Odyssey-3, a big step forward for foundation world models.\n\nIt can control robots, power humanoids, drive cars (on the roads of India!), train AIs, pilot drones, and even play video games.\n\nWe can&#8217;t wait to see what intelligent systems it enables. &quot;,&quot;username&quot;:&quot;odysseyml&quot;,&quot;name&quot;:&quot;Odyssey&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2071642815944359936/IebkUQ-C_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-15T16:36:04.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!5iDQ!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2099897328866115585.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/iNBWkPTfOO&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:258,&quot;retweet_count&quot;:535,&quot;like_count&quot;:4415,&quot;impression_count&quot;:2258361,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2099897328866115585/vid/avc1/1280x720/ib8AypH9KFG47XQh.mp4&quot;,&quot;video_preview_media_key&quot;:&quot;13_2099897328866115585&quot;,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong>Hardware founders are multiplying. </strong>A founder can now describe a part and get usable CAD, have a board fabricated and assembled for &lt;$100, buy a quadruped with a full SDK for $2,500, fine-tune an open-sourced policy instead of training one, and pull the first hundred units from an overseas supplier without leaving their desk. What previously took a year and $250K now takes a few weeks and a credit card. Prototyping was never what killed hardware companies; going from 10 units to 1,000 is, and that takes production tooling, safety certification, field service and maintenance, and working capital. We think cheap prototyping increases the number of companies that reach that wall rather than the number that clear it, which is precisely why we keep looking at the financing and insurance layers.</p><div><hr></div><p><strong>AI decisions are about to get too cheap to meter.</strong> <a href="https://x.com/CompleteSkeptic">Diogo Almeida</a>, who worked on the RLHF research behind ChatGPT, <a href="https://x.com/CompleteSkeptic/status/2099925682726002904">came out of stealth</a> with TypeSafe and a model called Jev that doesn&#8217;t generate text at all. It takes unstructured state and returns typed outputs with calibrated probabilities, at $0.042 per million input tokens with output free, in 70 to 500 milliseconds. The evals are self-reported and the reference answers come from frontier models, so discount the 200x claims accordingly. This matters because most production AI work isn&#8217;t conversation, it&#8217;s a fuzzy if-statement buried inside software, and nobody has been pricing it as its own primitive. We think the interesting question for founders is whether this is a category or a feature the labs absorb within a year.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/CompleteSkeptic/status/2099925682726002904&quot;,&quot;full_text&quot;:&quot;After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?\n\nI&#8217;ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev\n\n&#8226; 20-200x faster\n&#8226; 40-400x &#8230;&quot;,&quot;username&quot;:&quot;CompleteSkeptic&quot;,&quot;name&quot;:&quot;Diogo Almeida&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1650708125685800960/7k6r0UZg_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-15T18:17:52.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!z-75!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2099925575637057536.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/JSybNG2BKJ&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:3509,&quot;retweet_count&quot;:7023,&quot;like_count&quot;:66618,&quot;impression_count&quot;:31687933,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2099925575637057536/vid/avc1/1280x720/ZoKJ_BS5SNaqG45k.mp4&quot;,&quot;video_preview_media_key&quot;:&quot;13_2099925575637057536&quot;,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong>Why do agents lie?</strong> After the OpenAI-Hugging Face incident, Yoshua Bengio <a href="https://x.com/yoshua_bengio/status/2098419464597295145">published</a> the clearest account we&#8217;ve read of why misaligned behavior keeps recurring: safety goals are vague, task goals are sharp, and a capable optimizer finds the reading of the vague one that lets it win the sharp one. His uncomfortable corollary is that better monitoring may just select for agents that cheat without getting caught. This matters because it reframes agent safety as a permanent operating cost rather than a bug to be patched. We think it&#8217;s the strongest case yet for the verification stack, meaning evals, runtime monitors, permissioning, audit trails, and eventually insurance, because once agents take actions with financial consequences, somebody has to underwrite those consequences.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/yoshua_bengio/status/2098419464597295145?s=52&quot;,&quot;full_text&quot;:&quot;Over the past few days, I've taken the time to summarize my thoughts on the recent incidents involving agents&#8217; misaligned behavior. We don't know with certainty what comes next, but we know where these issues originate, and this can help us plan the path forward.\n\nPlease feel &#8230;&quot;,&quot;username&quot;:&quot;Yoshua_Bengio&quot;,&quot;name&quot;:&quot;Yoshua Bengio&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2033915590566428672/NVNxLW6g_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-11T14:32:41.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HR8VW-oWMAcfXsk.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/BYBAySE0Cc&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:166,&quot;retweet_count&quot;:494,&quot;like_count&quot;:2223,&quot;impression_count&quot;:433089,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong>Mapping the AI buildout, site by site.</strong> Epoch AI <a href="https://x.com/epochairesearch/status/2100278844456599798">expanded its data center research</a> to 86 sites covering 44% of global AI compute and 53% of everything deployed this year, with satellite imagery, capacity in H100-equivalents, chip types, capital costs, and construction status for each. Chinese coverage sits at 9-31%, which is its own story. This matters because the largest infrastructure buildout in history has been financed almost entirely on private information, and site-level data on capacity and construction is the beginning of a real underwriting dataset. We think you can&#8217;t price collateral you can&#8217;t see, and the lenders, insurers, and infrastructure debt markets now moving into compute all need to see it.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/epochairesearch/status/2100278844456599798?s=46&quot;,&quot;full_text&quot;:&quot;We're scaling our AI Data Centers research to provide a more comprehensive map of the world's AI infrastructure. Our data now captures ~44% of global AI compute, with coverage growing fast. &quot;,&quot;username&quot;:&quot;EpochAIResearch&quot;,&quot;name&quot;:&quot;Epoch AI&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1866142753127616512/DYcE9bN1_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-16T17:41:12.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HSWwepbbAAAUgiF.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/NJpD2pJajx&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:3,&quot;retweet_count&quot;:15,&quot;like_count&quot;:111,&quot;impression_count&quot;:18673,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><em>We&#8217;ll share another edition next week.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Who Underwrites the Robots?]]></title><description><![CDATA[Robotics startups are buying fleets with venture equity. They should be borrowing.]]></description><link>https://blog.canonical.cc/p/who-underwrites-the-robots</link><guid isPermaLink="false">https://blog.canonical.cc/p/who-underwrites-the-robots</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 11 Sep 2026 12:01:53 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/2f699460-67cc-4d47-ac86-8e95151fd6ad_1672x941.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><span>Roughly 25% of the companies we&#8217;ve looked at this year have been robotics companies. Nearly all of them, from pre-seed to Series A, raise venture equity and spend a big chunk of it on hardware, including the fleets they deploy to customers.</span></p><p><span>In most industries that would be strange. Trucking companies don&#8217;t sell equity to buy trucks, and farmers don&#8217;t sell stakes in the farm to buy tractors. Machines that earn contracted revenue get bought with debt, because a lender can underwrite the cash flows and take the machine back if they stop. Robotics startups are funding exactly that kind of asset with the most expensive capital they&#8217;ll ever raise.</span></p><h2><span>Equity is buying the fleet</span></h2><p><span>A commercial robot costs $20-100K. Under a signed service or lease contract, it earns a few thousand dollars a month and pays for itself in 12-18 months. Land a customer for 50 units and you need millions in hardware before revenue arrives. Most startups can only fund that with another equity round, taking permanent dilution to cover an 18-month gap. Robotics startups </span><a href="https://news.crunchbase.com/robotics/startup-venture-funding-surges-2026-data/"><span>raised $18.8B globally by June</span></a><span>, more than in all of 2025, and a lot of it is buying hardware. </span><strong><span>One of our portfolio companies told us they&#8217;d put $20M into their fleet tomorrow if they could get it, but raising it as equity at their stage isn&#8217;t realistic.</span></strong></p><p>In theory, there are three ways to fund this without equity:</p><ol><li><p><strong>Equipment financing</strong> covers the machines used to build robots, like CNC machines, which have deep resale markets.</p></li><li><p><strong>Fleet financing</strong> lends against robots a startup deploys itself under a signed contract. The lender is repaid from the contract and can repossess the robots if that fails.</p></li><li><p><strong>Customer financing</strong> does the most. A lender pays the startup upfront and the customer repays over its contract. The startup gets cash on delivery to buy its next batch of parts, and the credit risk moves to the customer, who is often a much bigger company.</p></li></ol><p>The parts themselves are the hardest to lend against. Motors and gears get customized, built into prototypes, or used up in R&amp;D, and their resale value drops fast once that happens.</p><h2><span>Why nobody lends</span></h2><p><span>Banks want years of operating history and profits. Equipment lessors want two years of financials, which is unrealistic for a seed-stage company. Venture debt usually starts at Series B, comes with warrants, and takes a lien on the whole company, so the founder pledges everything to finance one fleet.</span></p><p><span>Bank capital rules make it worse, and </span><a href="https://usd.ai/insights/banks-finance-gpu-credit"><span>GPUs show how</span></a><span>. Under Basel III, a loan repaid from a specific asset&#8217;s cash flows counts as &#8220;object finance&#8221; and gets slotted into preset risk buckets that punish anything unproven. So banks lend to private credit funds instead, and the funds lend against the GPUs at higher rates.</span></p><p><a href="https://usd.ai"><span>USD.AI</span></a><span> was built for that gap. It takes stablecoin deposits and lends them against tokenized GPUs. Its largest loan grew from $620K in July 2025 to $98.1M this June, with </span><a href="https://usd.ai/insights/usdai-2026-ytd-report-lighthouse"><span>$202M deployed and ~$30M of annualized revenue in Q2</span></a><span>. Deposits went the other way, from $650M+ at its January token launch to </span><a href="https://www.coindesk.com/business/2026/08/28/bullish-backs-usd-ai-with-usd100-million-gpu-stablecoin-financing"><span>~$225M by August</span></a><span>. On-chain capital is easy to attract. Origination is the business.</span></p><h2><span>Insurance is moving faster</span></h2><p><a href="https://www.corgi.insure"><span>Corgi</span></a><span> is an AI-native insurance carrier for startups. Instead of acting as a broker, it writes and prices policies itself, using company data rather than long application forms. It was founded in 2024, reported $40M of annualized revenue at its January Series A, and </span><a href="https://techcrunch.com/2026/07/23/insurance-startup-corgi-reportedly-raised-more-money-at-4b-its-third-round-in-eight-weeks/"><span>reportedly raised at a $4B valuation in July</span></a><span>, tracking toward a $450M run rate by year end. The lesson is that a carrier underwriting from data can win the categories incumbents handle badly, and robots are one of them.</span></p><p><a href="https://www.trustboop.com/"><span>Boop</span></a><span> is already there. It sells per-robot insurance to companies deploying robots and includes a black-box recorder that settles claims on evidence. It argues standard liability policies likely exclude autonomous failures. Boop started out pitching insurance for household humanoids, and its pivot to commercial fleets shows where the demand is.</span></p><h2><span>The missing piece</span></h2><p><span>Robotics operators need financing and insurance at the same time, for the same robots. Both are underwritten on the same data: what the robot is, where it works, how much it&#8217;s used, and who&#8217;s paying for it.</span></p><p><span>Other machine industries solved this at the point of sale. When you buy a car, the dealer offers you a loan and insurance at the same desk. GM started its own lender, GMAC, in 1919 because customers couldn&#8217;t buy cars they couldn&#8217;t finance. John Deere&#8217;s lending arm </span><a href="https://equipmentfinancenews.com/news/lender-operations/john-deere-financial-earnings-rise-on-wider-spreads/"><span>holds ~$49B of loans today</span></a><span>.</span></p><p><span>Robotics doesn&#8217;t have this yet. Robot makers are young and don&#8217;t lend to their customers. We think a startup will fill that gap by offering financing and insurance right where robots are bought.</span></p><p><span>That&#8217;s where crypto comes in. Stablecoin holders want real yield, and short, contract-backed loans against working robots fit that well. But as USD.AI shows, capital isn&#8217;t the bottleneck. The hard part is originating good loans.</span></p><p><span>We&#8217;ve been thinking about this all year and recently met founders already doing that work. If you&#8217;re building credit or insurance for the physical economy, or you&#8217;re a robotics founder spending your seed round on a fleet, we&#8217;d like to talk.</span></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[1 Million AI Residents Need Passports]]></title><description><![CDATA[Simulated cities, critical cyber capability, NVIDIA&#8217;s custom-silicon strategy, generative worlds, and the identity layer for autonomous agents.]]></description><link>https://blog.canonical.cc/p/1-million-ai-residents-need-passports</link><guid isPermaLink="false">https://blog.canonical.cc/p/1-million-ai-residents-need-passports</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 04 Sep 2026 12:57:51 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/3b7ff55a-5fa4-452d-8fd7-93a36d3e5b7e_1536x1152.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Five ideas, papers, and products shaping our investment thinking, and what we think they mean.</em> <em>This series is powered by an AI assistant that helps synthesize recurring themes from our discussions, alongside our own reflections.</em></p><div><hr></div><p><strong><a href="https://arxiv.org/html/2506.21805v1"><span>Cities are software products.</span></a></strong> CitySim gives LLM agents beliefs, long-term goals, habits and spatial memory, then asks them to generate their own schedules. In a Tokyo setting, the system reproduced patterns in Japanese time-use and travel data, while forecasting Shibuya crowding and venue popularity better than prior agent systems. The paper demonstrates realistic behavior at 1,000 agents and <strong>system scalability to one million.</strong> Urban planning, site selection and infrastructure design can become iterative software workflows.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/thesupermanmx/status/2078847665752957086&quot;,&quot;full_text&quot;:&quot;Japanese researchers created a system that simulates an entire city by generating up to 1 million virtual residents who behave like humans using LLMs.\n\nAnd it predicted real-world with terrifying accuracy.\n\nThey created an urban simulator called \&quot;CitySim\&quot; that populates a digital &#8230;&quot;,&quot;username&quot;:&quot;thesupermanmx&quot;,&quot;name&quot;:&quot;Superman&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2061753751560650752/4jxd2vjA_normal.jpg&quot;,&quot;date&quot;:&quot;2026-07-19T14:21:21.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://pbs.substack.com/media/HNmM4eHbkAALOxW.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/yCBZFmWh3C&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:112,&quot;retweet_count&quot;:659,&quot;like_count&quot;:2485,&quot;impression_count&quot;:192328,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://openai.com/index/path-to-astra/"><span>AI will find more bugs than teams can fix.</span></a></strong> OpenAI classified <a href="https://openai.com/index/gpt-6-astra/">GPT-6 Astra</a> as Critical for cybersecurity after testing showed sharp advances in vulnerability discovery and exploit development. That means security teams will receive more potential issues than they can assess and resolve manually. We think winning products will identify which reports are real and urgent, generate patches, test them against production dependencies, and safely deploy or roll them back. We also think winning companies will pair autonomous remediation agents with human ownership and clear operational controls.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/markchen90/status/2095597534412673109&quot;,&quot;full_text&quot;:&quot;GPT-6 Astra is here! This is a big moment for our research team - years of work on pretraining, reinforcement learning, and post-training have come together in our most capable and aligned model yet. It can build and test software, work across apps on your computer, and even help&#8230;&quot;,&quot;username&quot;:&quot;markchen90&quot;,&quot;name&quot;:&quot;Mark Chen&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1273327644436656129/uYf1oNS5_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-03T19:39:21.000Z&quot;,&quot;photos&quot;:[],&quot;quoted_tweet&quot;:{&quot;full_text&quot;:&quot;This is GPT-6 Astra.\n\nAnything you can do on a computer, Astra can do for you. Fast.&quot;,&quot;username&quot;:&quot;OpenAI&quot;,&quot;name&quot;:&quot;OpenAI&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1885410181409820672/ztsaR0JW_normal.jpg&quot;},&quot;reply_count&quot;:77,&quot;retweet_count&quot;:122,&quot;like_count&quot;:1923,&quot;impression_count&quot;:371072,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:false}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://nvidianews.nvidia.com/news/nvidia-and-mediatek-deepen-long-standing-partnership-to-build-ai-edge-to-cloud-computing-platforms"><span>NVIDIA is making custom chips work inside its ecosystem.</span></a></strong> It invested $3.5B in MediaTek and will let MediaTek offer customers a way to connect custom AI accelerators to NVIDIA&#8217;s NVLink systems. Model labs and hyperscalers increasingly want silicon built around their specific workloads. But a chip alone is not a data center: it needs memory, networking, software, cooling and operational support that work together. This is a bullish NVIDIA story. Custom chips expand, but NVIDIA can still own the systems that deploy them at scale. Startups need a clear workload-specific advantage, not another generic accelerator.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/BloombergTV/status/2094773096817328180&quot;,&quot;full_text&quot;:&quot;Nvidia is making a $3.5 billion investment in MediaTek &#8212; and striking a deeper technology partnership around custom AI chips. CEO Jensen Huang told <span class=\&quot;tweet-fake-link\&quot;>@EdLudlow</span> that Nvidia isn&#8217;t threatened by the rise of specialized XPUs. Instead, it wants to make it easier for those chips to &#8230;&quot;,&quot;username&quot;:&quot;BloombergTV&quot;,&quot;name&quot;:&quot;Bloomberg TV&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1631751904475709441/fhQng3ic_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-01T13:03:19.000Z&quot;,&quot;photos&quot;:[{&quot;img_url&quot;:&quot;https://substackcdn.com/image/fetch/$s_!w7GQ!,w_1028,c_limit,f_auto,q_auto:best,fl_progressive:steep/l_play_button_usfui2,w_88,e_colorize:0/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F__ss-rehost__tw-video-preview-13_2094773065557192704.jpg&quot;,&quot;link_url&quot;:&quot;https://t.co/syYGeqVJKg&quot;}],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:3,&quot;retweet_count&quot;:10,&quot;like_count&quot;:29,&quot;impression_count&quot;:12123,&quot;expanded_url&quot;:null,&quot;video_url&quot;:&quot;https://video.twimg.com/amplify_video/2094773065557192704/vid/avc1/720x1280/7nHjHasWiyp-NE9h.mp4?tag=14&quot;,&quot;video_preview_media_key&quot;:&quot;13_2094773065557192704&quot;,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.diiverge.co/"><span>The next social network may be a world, not a feed.</span></a></strong> Diiverge turns a single image into a persistent point-and-click adventure. Choose an object, decide what happens, and the scene continues as video. Each choice creates a branch that remains available to the next visitor. This makes generative video feel less like a format for watching and more like an environment people can shape together. The interesting products will not simply generate infinite content. They will give people reasons to return to worlds they helped create, explore the paths other people made, and build a shared history inside them.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/rehan_shei/status/2095447542448206012&quot;,&quot;full_text&quot;:&quot;For those wondering how he did this:\n\n1. Generate vid with H3 Max/Max Turbo\n2. Extract last frame run Sam3 to extract object masks\n3. Prompt LLM for continuation prompts for top 3 objects\n4. Generate all N options per object using H3 Max/Max Turbo I2V with image ref set as last&quot;,&quot;username&quot;:&quot;rehan_shei&quot;,&quot;name&quot;:&quot;Rehan Sheikh&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1836900265959772161/tuQKDoZ6_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-03T09:43:20.000Z&quot;,&quot;photos&quot;:[],&quot;quoted_tweet&quot;:{&quot;full_text&quot;:&quot;Aaaaaaand we're live!\n\nhttps://t.co/mrHuwmjq8T\n\nA persistent, infinite, point-and-click adventure.\n\nEvery picture is a fork. Click something in it, decide what happens, and the world grows in that direction. https://t.co/7iG8AqHick&quot;,&quot;username&quot;:&quot;charliie&quot;,&quot;name&quot;:&quot;Charlie Clark&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/2015454175203573761/AXFfStJu_normal.jpg&quot;},&quot;reply_count&quot;:14,&quot;retweet_count&quot;:43,&quot;like_count&quot;:894,&quot;impression_count&quot;:77497,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><strong><a href="https://www.hyperdimensional.co/p/on-the-loose"><span>Agents need identity before they can become economic actors.</span></a></strong> Long-running agents will hold credentials, purchase compute, call services, move money and build reputations over time. That requires persistent identity, clear permissions and accountability when something goes wrong. The debate around self-sovereign agents is still speculative, but the infrastructure problem is already real. An agent acting for a person, an organization or itself needs different rights and limits. We think agent identity becomes a core layer of the agentic economy, making autonomy legible to businesses, financial institutions and the systems agents will increasingly operate.</p><div class="twitter-embed" data-attrs="{&quot;url&quot;:&quot;https://x.com/deanwball/status/2094794694572015854?s=20&quot;,&quot;full_text&quot;:&quot;https://t.co/dv55KyXDQR&quot;,&quot;username&quot;:&quot;deanwball&quot;,&quot;name&quot;:&quot;Dean W. Ball&quot;,&quot;profile_image_url&quot;:&quot;https://pbs.substack.com/profile_images/1997065021491130368/X76ALSbp_normal.jpg&quot;,&quot;date&quot;:&quot;2026-09-01T14:29:09.000Z&quot;,&quot;photos&quot;:[],&quot;quoted_tweet&quot;:{},&quot;reply_count&quot;:53,&quot;retweet_count&quot;:66,&quot;like_count&quot;:455,&quot;impression_count&quot;:133044,&quot;expanded_url&quot;:null,&quot;video_url&quot;:null,&quot;video_preview_media_key&quot;:null,&quot;belowTheFold&quot;:true}" data-component-name="Twitter2ToDOM"></div><div><hr></div><p><em>We&#8217;ll share another edition next week.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[The Wrong Question About Robots]]></title><description><![CDATA[Why we invested in Robo]]></description><link>https://blog.canonical.cc/p/the-wrong-question-about-robots</link><guid isPermaLink="false">https://blog.canonical.cc/p/the-wrong-question-about-robots</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 21 Aug 2026 21:53:07 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/a6c02982-0206-4eb6-854b-95289085b892_1512x982.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>This week I was at 3 events in SF where the discussion centered on whether we&#8217;re in the GPT-2 or GPT-3 era in robotics. I think that&#8217;s the wrong question.</p><p>The underlying assumption here is that progress in robotics is predicated on step-function model breakthroughs. But models and task reliability are already improving consistently and scaling behavior looks real - this isn&#8217;t the bottleneck!</p><p>The real bottleneck is economics - we need cheap, globally scalable robots that can be put in a warehouse, hospital, or field and drive top-line improvements or efficiency gains. Robotics doesn&#8217;t need a GPT-3 moment. <a href="https://new.robo.inc/blog/robotics-model-t-moment">It needs a Model T moment.</a></p><h3><strong>Cost is the entry point</strong></h3><p>Cheap robots help make robotic deployments economically feasible and accessible. Robotic deployments create data, data improves autonomy, and autonomy lowers cost and increases value. <strong>We think this loop is the most important one in robotics.</strong></p><p>This flies in the face of traditional hardware economics, where companies treat cost as something to address only at scale. History teaches us to first build the most capable version, sell it at the top of the market, and use that to subsidize the ramp, driving BOM cost down later with economies of scale.</p><p>Unfortunately, that doesn&#8217;t work in robotics, because you need deployments to generate the data that improves autonomy, and you need better autonomy to increase value-per-robot and robot-per-teleoperator ratio.</p><p>In other words, a robot that starts expensive never gets enough deployments to become cheap. It gets stuck at the top of the loop.</p><h3><strong>Robo</strong></h3><p>Today, our portfolio company <a href="https://robo.inc/">Robo</a> comes out of stealth. Robo builds affordable, general-purpose robots and teleoperation infrastructure.</p><p>Their first robot is Robo-T. Designed and built here in Los Angeles, it holds 10 pounds and runs tasks across warehouses, hospitals, food manufacturing, and logistics environments.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!SUrl!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!SUrl!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 424w, https://substackcdn.com/image/fetch/$s_!SUrl!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 848w, https://substackcdn.com/image/fetch/$s_!SUrl!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 1272w, https://substackcdn.com/image/fetch/$s_!SUrl!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!SUrl!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg" width="378" height="504" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/dafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1600,&quot;width&quot;:1200,&quot;resizeWidth&quot;:378,&quot;bytes&quot;:337847,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/jpeg&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/212212161?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!SUrl!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 424w, https://substackcdn.com/image/fetch/$s_!SUrl!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 848w, https://substackcdn.com/image/fetch/$s_!SUrl!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 1272w, https://substackcdn.com/image/fetch/$s_!SUrl!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdafc44f5-5a43-4663-930b-aacc86829fb2_1200x1600.jpeg 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Over the past three months I&#8217;ve seen Robo go from robotic-effectors to robot arms to a full mobile embodiment, and relentlessly cut deployment costs by owning the full stack to just $10/hour. They also built <a href="https://robo.inc/#roboport">Roboport</a>, their own teleoperation platform, rather than renting one. It turns demonstrations, rollouts, and interventions into improving autonomy, helping close the cost-autonomy loop.</p><p>Robo&#8217;s founders, <a href="https://www.linkedin.com/in/don-morton/">Don</a> and <a href="https://www.linkedin.com/in/kylenoble/">Kyle</a>, are high-integrity, high-velocity builders with a rare blend of technical depth, product intuition, and customer obsession. They pivoted into robotics midway through our diligence, and seeing their speed and pragmatism helped get us over the line at pre-seed.</p><p>Robo is now looking for fleet partners. Similar to operators of Tesla fleets for Uber or Lyft - Robo are looking for people who buy 15+ Robo-Ts, deployed around the clock, and sell the hours. Those fleets ended up providing much of the liquidity on Uber&#8217;s supply side, and turned out to be very good businesses for the people running them. Robo handles delivery, setup, and maintenance; what you need is organization and the communication skills to teach operators a task.</p><p>If you&#8217;re building here, or want to run a fleet, <a href="https://robo.inc/contact"><span>reach out to them directly</span></a>. And if you&#8217;re working on anything at the intersection of cheap embodiment and real deployment, <a href="mailto:hello@canonical.cc"><span>come talk to us</span></a> at Canonical.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Synthefy: Tabular Foundation Models Will Matter More to Enterprises Than LLMs]]></title><description><![CDATA[Beyond Text. Why the Market for "Number Tokens" Will Outgrow LLMs]]></description><link>https://blog.canonical.cc/p/synthefy-timeseries-foundation-models</link><guid isPermaLink="false">https://blog.canonical.cc/p/synthefy-timeseries-foundation-models</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Tue, 18 Aug 2026 17:39:54 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zAf9!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>I met <a href="https://www.linkedin.com/in/shubhankar-agarwal/">Somi</a> back in 2023, when he was wrapping up his PhD at UT Austin under Prof. Sandeep Chinchali. There was no company but the initiative they were experimenting with was called &#8220;GeneSys&#8221;, and they were using AI to test complex, high-stakes production systems. </p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!zAf9!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!zAf9!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 424w, https://substackcdn.com/image/fetch/$s_!zAf9!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 848w, https://substackcdn.com/image/fetch/$s_!zAf9!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 1272w, https://substackcdn.com/image/fetch/$s_!zAf9!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!zAf9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg" width="1456" height="1092" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1092,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:2850488,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/jpeg&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/211733333?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!zAf9!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 424w, https://substackcdn.com/image/fetch/$s_!zAf9!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 848w, https://substackcdn.com/image/fetch/$s_!zAf9!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 1272w, https://substackcdn.com/image/fetch/$s_!zAf9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a8fda82-2e1d-4de6-a38b-618afc9ede1e_2390x1792.jpeg 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a><figcaption class="image-caption">Somi presenting the first avatar of Synthefy at the Summer Lab UT Austin, June 2023</figcaption></figure></div><p>Over the next few months I spent a lot of time with Somi and Sandeep as the vision for <a href="https://www.synthefy.com/">Synthefy</a> came together. Later that year, Canonical had written the first check as the only investor in their pre-seed, and we started meeting every week.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>The idea they landed on was simple, driven by a core thesis: </p><div class="callout-block" data-callout="true"><p><strong>the market for &#8220;number tokens&#8221; is bigger than text tokens. </strong></p></div><p>Most of the data that runs the economy lives in tables and time series: transactions, prices, sensor readings, inventory, customer histories. Predictions made from that data decide which payments get approved, how products get priced and stocked, which customers leave, and when machines fail.</p><p>And the way companies make those predictions has barely changed in a decade. Every workflow is a custom project: clean the data, engineer features, train a model, tune it, deploy it, maintain it forever. The Synthefy team lived this at Uber&#8217;s self-driving unit, where dozens of people spent 5-6 months building a single model. A production system can cost $500K-$1.5M to build and up to $750K a year to keep running. </p><blockquote><p>The team believed pretraining would do for numbers what it did for language: one model, trained on huge amounts of structured data, that makes accurate predictions on a table it has never seen. Six months of ML work becomes an API call.</p></blockquote><p>The path from there looked like most good pre-seed paths, which is to say &#8216;quiet&#8217;. First MVP in early 2024, pointed at defense and telecom workloads. Deutsche Telekom named them best AI startup.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!zFDH!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!zFDH!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 424w, https://substackcdn.com/image/fetch/$s_!zFDH!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 848w, https://substackcdn.com/image/fetch/$s_!zFDH!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 1272w, https://substackcdn.com/image/fetch/$s_!zFDH!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!zFDH!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png" width="1314" height="1448" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1448,&quot;width&quot;:1314,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:2316489,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/211733333?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb039d9be-3e43-4aba-806c-a91e7a322d64_1314x2480.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!zFDH!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 424w, https://substackcdn.com/image/fetch/$s_!zFDH!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 848w, https://substackcdn.com/image/fetch/$s_!zFDH!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 1272w, https://substackcdn.com/image/fetch/$s_!zFDH!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F181f6242-5901-46dd-b511-55edfb77baf8_1314x1448.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Through 2025 Synthefy stayed heads down: deployments at enterprises like GoodRx, Schneider Electric, and a US Army contract, while the rest of the market has been chasing text, images &amp; videos through LLMs.</p><p>Then the market turned, fast. In May, SAP agreed to acquire Prior Labs, an 18-month-old German lab that had raised &#8364;9M, and committed over &#8364;1B to build on it. In June, Nvidia bought Kumo for a reported $400M. Two of the largest companies on earth paid up for the category within 30 days of each other.</p><p>Weeks later, Synthefy shipped Nori. It ranks #1 for accuracy on a public benchmark of 96 datasets, ahead of Google&#8217;s TabFM and the TabPFN model SAP had just bought and remarkably, it does it with just <strong>6M parameters, a tenth the size of comparable models</strong>. It crossed 440K downloads in its first 30 days, and <strong>recently surpassed 630K downloads</strong>. </p><p>In pilots, it beat forecasting systems customers had spent years tuning, with no custom training. </p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!7T8p!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!7T8p!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 424w, https://substackcdn.com/image/fetch/$s_!7T8p!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 848w, https://substackcdn.com/image/fetch/$s_!7T8p!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 1272w, https://substackcdn.com/image/fetch/$s_!7T8p!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!7T8p!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png" width="1456" height="819" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/ada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:819,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1898166,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/211733333?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!7T8p!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 424w, https://substackcdn.com/image/fetch/$s_!7T8p!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 848w, https://substackcdn.com/image/fetch/$s_!7T8p!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 1272w, https://substackcdn.com/image/fetch/$s_!7T8p!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fada3486f-cd43-45d8-a924-11a6856ed217_3840x2160.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Today Synthefy announces its seed round, led by <a href="https://www.wing.vc/">Wing</a> with <a href="https://haystack.vc/">Haystack</a>, Canonical, <a href="https://www.samsungnext.com/">Samsung Next</a>, plus angels including <a href="https://www.linkedin.com/in/srinivasnarayanan/">Srinivas Narayanan (ex-CTO at OpenAI)</a>, <a href="https://www.linkedin.com/in/aparnacd/">Aparna Chennapragada (CPO at Microsoft)</a>, and <a href="https://www.linkedin.com/in/manoharpaluri/">Manohar Paluri (VP of AI at Meta)</a>.</p><p>Congrats, Somi and team. The economy&#8217;s numbers finally have their foundation model!</p><p>PS: One note on sourcing, because people ask how a team like ours finds Somi and Sandeep. The best pre-seed deals rarely look like &#8220;deals.&#8221; They look like research: no deck, no go-to-market plan, sometimes no graduation date. Our job is to be in the lab while the conversation is still &#8220;could this be a company,&#8221; and to be useful.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Ways to Win in Robotics]]></title><description><![CDATA[The market is too big for one winner - enduring companies will either own the workflow or open the stack.]]></description><link>https://blog.canonical.cc/p/ways-to-win-in-robotics</link><guid isPermaLink="false">https://blog.canonical.cc/p/ways-to-win-in-robotics</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Sat, 15 Aug 2026 00:10:51 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/7d0ec081-6106-453d-bae1-a1f4a8145018_1536x1152.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Software investors have been taught a simple mental model on where value accrues: </p><ol><li><p>companies building deep infrastructure at the bottom of the stack that is hard to replicate, e.g. NVIDIA, AWS, Cloudflare, Snowflake</p></li><li><p>companies building applications with incredible distribution at the top of the stack, e.g. Facebook, Instagram, OpenAI, Figma</p></li></ol><p>Everything in the middle gets compressed.</p><p>I think that model breaks down in robotics - it&#8217;s just way too big and too broad for single winners! Instead, we&#8217;ll see winning robotics companies in mining, agriculture, logistics, deep sea, defense, manufacturing, healthcare, and dozens of other verticals. These companies will win because they operate in different environments with different safety requirements, regulations, buyers, economics, and failure modes. A winning underwater robot is likely to be irrelevant in a hospital setting. A robot designed to pick and pack fragile inventory stock will be irrelevant when maintaining a solar farm.</p><p>With this in mind, I think robotics will produce many large companies, rather than one dominant infrastructure or application layer.</p><p>I expect those companies to broadly fall into two categories:</p><ol><li><p>vertically integrated companies <strong>that own a specific workflow end-to-end</strong>, and</p></li><li><p>companies that benefit from <strong>the modularization of the robotics stack</strong>.</p></li></ol><h3><strong>1/ Vertically Integrated Companies</strong></h3><p>Think of these companies as taking Apple&#8217;s path in robotics: increasingly owning the hardware, software, models, deployment, and customer relationship around one valuable physical job. What&#8217;s most important for these companies is <strong>owning the operating loop.</strong></p><p>Take, for example, a robotic mining company - they may start with third-party vehicles, NVIDIA compute, open models, and off-the-shelf sensors. But, so long as they own the operating loop, every hour teaches that company something a generic robotics company cannot know. They learn about how dust affects perception, how different routes change tire wear, which slopes create control problems, and which component failures produce downtime.</p><p><strong>That knowledge improves the system.</strong> It improves the policies, maintenance schedules, sensor configurations, fleet-management software, and customer workflow. Over time, it also tells the company which hardware should be customized, built, or acquired. <strong>The company earns the right to build custom hardware because it owns the job.</strong></p><p>I think that same pattern will appear everywhere physical work is valuable and technical. For example: an agricultural robotics company will learn which weeds emerge at each crop stage, how weather and light affect vision, which treatments work, and where its equipment loses performance. A subsea maintenance company learns about marine conditions, asset degradation, battery performance, and the specific equipment customers need inspected or repaired. A logistics company learns how a facility actually operates: where inventory sits, when workers intervene, what causes exceptions, and which tasks create the greatest bottlenecks.</p><p>This deployment data makes the system more useful and more economical over time.</p><p>This gives those companies a deep competitive advantage, <strong>enabling them to sell outcomes, i.e. tonnes moved or packages packed, and not just machines or hours.</strong> They may start with third-party components, but the operating loop gives them a growing advantage over generic robot OEMs.</p><h3><strong>2/ Modular Platform Companies</strong></h3><p>The winners in this category will be the companies that make it dramatically easier for other people to build, deploy, and operate robots.</p><p>Rather than owning a single workflow, such as mining or agriculture, <strong>they will provide the shared infrastructure that many robotics companies rely on:</strong> fleet management, remote operations, simulation, data infrastructure, safety tooling, hardware interfaces, and marketplaces for task-specific models and components.</p><p>Their opportunity comes from the robotics stack becoming more modular. Hardware will get cheaper. Open models will become good enough for more tasks. Components will become more interoperable. And new platforms will let teams deploy robot fleets without rebuilding every layer of the stack from scratch.</p><p>A company should increasingly be able to combine a low-cost robot body from one vendor, a specialized gripper from another, an open model, and a shared fleet-management layer, then deploy the system into a real workflow. That is a much more accessible world than one where every robotics company has to build its own hardware, models, tooling, and operating software.</p><p>This is what open systems do well. They lower the cost of experimentation, bring in more builders, and speed up the entire market.</p><p>The &#8220;Android for robotics&#8221; opportunity is real.</p><p>But open source alone is not a business model. Open models and interoperable hardware will drive adoption, but the best platform companies will own <strong>a control point that compounds</strong>, e.g. the default fleet-management layer for heterogeneous robotics or managed data and simulation infrastructure. The key question for builders and investors alike is <strong>whether every new deployed fleet makes the platform better.</strong></p><p>A company managing thousands of robots across different environments will learn which components work together, where deployments fail, and how to reduce the time and cost of getting a robot into production. Those insights improve the platform, helping attract more builders and fleets, helping further improve the platform.</p><p>This is a different type of operating loop from the vertically integrated company. The vertical company learns how to perform one specific job better than anyone else. The modular platform company learns how to help many different companies deploy robots more effectively.</p><p><strong>Both can create enormous businesses.</strong></p><p>At Canonical, we are particularly excited by vertically integrated companies with proprietary access to difficult physical workflows that improve with every deployment. We also want to meet teams building the modular platforms that help the broader robotics ecosystem build and deploy faster.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[ The Open-Weight AI Debate]]></title><description><![CDATA[The open-weights fight isn't really about openness. It's about which danger arrives first, and which one founders can build against today.]]></description><link>https://blog.canonical.cc/p/the-open-weight-ai-debate</link><guid isPermaLink="false">https://blog.canonical.cc/p/the-open-weight-ai-debate</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Sat, 01 Aug 2026 00:53:16 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/0b419b01-80a0-40bd-b972-613f69face74_1088x608.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>The open-weights debate escalated this week, turning into something more useful than another round of &#8220;freedom vs. safety.&#8221;</p><p>Zuck made <a href="https://www.wsj.com/opinion/the-ai-future-is-for-everyone-a0c24e20">the political case</a> for open-weight AI, an NVIDIA-led coalition made <a href="https://images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf">the economic case</a>, and Illia made <a href="https://ilblackdragon.substack.com/p/why-pacing-the-ai-frontier-will-not">the systems case</a>. Meanwhile, 1,324 employees of frontier AI companies made <a href="https://www.pacingthefrontier.com/">the case for restraint</a>.</p><p>What divides them is which danger comes first: a few companies owning everything, America falling behind China, AI finding software holes faster than anyone can patch them, or models improving themselves beyond anyone&#8217;s ability to follow.</p><div><hr></div><p>Zuck&#8217;s case is about power: no single superintelligence can represent everyone&#8217;s values, so access must be distributed. Concentrated intelligence produces concentrated authority. That makes sense, but it also assumes equal compute, which remains unequal. Whoever can afford more inference, better memory, lower latency and more agents still has the stronger army. Weights can become open while power stays scarce.</p><p>Zuck then concedes that biological risks may require coordinated limits on how capable models are deployed. Open weights make those limits harder to enforce. Once the weights are released, anyone can run or modify them, and the original developer cannot take them back. His answer to concentrated power leaves this irreversibility problem unresolved.</p><div><hr></div><p>NVIDIA&#8217;s case is that America will not win by producing a single frontier model. It wins by spreading American models globally before China does. Open weights lower costs, prevent lock-in and let organizations run AI on their own infrastructure.</p><p>The underlying incentive makes sense for NVIDIA. It wins when the model layer becomes cheap and compute remains scarce. Open weights push models toward commodity economics, while every fine-tune, evaluation and deployment still burns chips and energy. America gets broader distribution of American AI. NVIDIA gets a market where nobody upstream captures all the margin.</p><p>That incentive does not invalidate the economic case. It explains why NVIDIA is leading it.</p><div><hr></div><p>The 1,324 frontier AI employees worry that models will soon improve AI research itself, creating a feedback loop that moves faster than labs can understand or control. The labs are asking for tools and governance that let society control the pace before that happens.</p><p>Their strongest objection to open weights is irreversibility. A hosted model can be monitored, restricted or withdrawn. Released weights cannot.</p><p>Anthropic makes this case harder to dismiss because its position is narrower than &#8220;open models are dangerous.&#8221; Amodei has called <a href="https://www.anthropic.com/news/position-open-weights-models">lower-capability open models a public good</a>. He objects when downloadable weights become powerful enough to cause catastrophic harm and can no longer be monitored or recalled. This makes sense, but is also self-interested: restrictions on powerful open weights give frontier labs more time to extend their lead.</p><p>The open side still has the strongest empirical response. Open-weight models have been around for years, and no catastrophe has arrived. Open models have enabled independent research, local deployment, specialized models and competition. They have given startups an alternative to frontier-model APIs and governments a way to run sovereign AI.</p><p>That record matters. Restrictors have to explain what changes at the next capability threshold, where that threshold sits and why the record so far no longer applies. Their answer is simple: yesterday&#8217;s models could be safe to release while tomorrow&#8217;s are not.</p><div><hr></div><p>Finally, Illia agrees that more capable models create new risks. He disagrees that restricting access to their weights solves them.</p><p>His point is that open vs. closed tells us who can access a model, not whether the system running it is secure. A hosted model can still access sensitive data, call the wrong tools or exploit vulnerable software. Securing AI therefore means securing the whole stack: the model, data, execution environment, hardware and surrounding software.</p><p>This is the strongest of the three open arguments. It is also incomplete. Better security can constrain models running inside systems we control. It cannot recall released weights or stop someone from running them elsewhere.</p><p>Illia&#8217;s position rests on a specific bet: security improves faster than dangerous models spread beyond its reach. If capability wins that race, Anthropic&#8217;s irreversibility objection stands. If security keeps pace, the open vs. closed distinction matters much less.</p><div><hr></div><p>We think all four dangers are real. They just run on different clocks. Concentration is here now. The race with China plays out over years. Recursive self-improvement may arrive at any point, or not at all. The security gap is the one founders can build against today. That makes it the investable theme we&#8217;re tracking.</p><p>If securing AI is about the whole stack rather than the weights alone, the buildable surface is concrete: sovereign deployment, independent evaluation and red-teaming, agent security, verifiable inference and formal verification.</p><p>Some of this already has buyers and budgets. Companies need to know which model ran, what data it accessed and which actions it took. Verifiable inference and formal verification are earlier, more expensive and less proven.</p><p>We&#8217;re looking to fund companies selling security today with a credible path toward becoming verification infrastructure later. The bet is that security improves fast enough to constrain dangerous models before they spread beyond its reach.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe to receive new posts.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Crypto: Usage Up, Price Down]]></title><description><![CDATA[The speculators left and the price fell. The usage they were leaving behind compounded the whole way down.]]></description><link>https://blog.canonical.cc/p/crypto-usage-up-price-down</link><guid isPermaLink="false">https://blog.canonical.cc/p/crypto-usage-up-price-down</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 24 Jul 2026 16:25:03 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/9fc32b48-60bb-4e42-9ec7-79f650f63f7b_1088x608.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Crypto had its worst quarter in years while its usage hit new records. Both things are true at the same time. Price and usage are two different measurements of the same industry, and this year they stopped telling the same story. Price tracks the marginal speculator, the last person willing to bid. Usage tracks the committed builder, the one who keeps shipping whether or not anyone is bidding. When sentiment turns, the speculator leaves and the price falls. The builder stays. A falling price and rising usage are not a contradiction, then. They are two different groups being counted at the same time, and only one of them left.</p><p>That gap is where an early-stage investor earns the return.</p><p>Through this drawdown the speculators walked and the price broke, while the infrastructure kept getting laid and used. Measure the industry against the 2022 bottom instead of against last year, and the picture inverts. Ethereum transaction activity runs roughly thirteen times its 2022 level. Stablecoin supply has roughly doubled to around $300 billion and held there straight through the selloff. Value locked in DeFi is up more than 60%. Tokenized real-world assets sit at a record near $33 billion, led by tokenized Treasuries. Same bear-market price, an industry twice the size. Usage compounded the entire way down. Only the price disagreed.</p><p>The clearest tell is where the market already votes with real money. Crypto equities are up about 23% this year while the tokens are down about 36%. That split is not noise. The market is paying for the crypto it can underwrite and declining to pay for the crypto it can only speculate on. Miners sitting on signed power contracts are redirecting that energy into AI compute and getting valued as infrastructure. Lending protocols with real revenue are getting valued as businesses; one of them cleared roughly $900 million in fees over the past year. For the first time the tooling exists to see this plainly, because the fees and the buybacks settle on-chain where anyone can audit them. Cash flow earns a multiple. Narrative earns a chart.</p><p>There is a deeper shift underneath that split. For a decade tokens traded at a premium to the businesses beneath them. Today many trade at a discount to the same cash flows. The market has stopped paying up for the idea of a network and started paying for the economics of one. That is not the category dying. It is the category growing up.</p><p>Real institutions are building through the winter rather than waiting for the thaw. Federal stablecoin rules finalize this year and take effect in January. Stripe, Visa, BlackRock, and Coinbase are standing up new stablecoins behind that framework. Schwab and E*Trade are adding retail crypto rails to tens of millions of accounts. None of that is a wager on next quarter&#8217;s price. It is capacity being poured while the ground is still cheap, by counterparties who do not build on sentiment.</p><p>This is where the early-stage job actually gets done. Bull markets reward trading. Bear markets reward underwriting. The founders laying rails and shipping product right now are doing it into flat charts and thin attention, which is exactly what makes them cheap to back and exactly the kind of people worth backing. Anyone can fund usage after the price has confirmed it. The return comes from funding it while the price still argues.</p><p>So the discipline is simple. Ignore the quote and read the usage, because the usage is the business and the quote is only the mood. Sentiment collapses in a quarter and the commitment underneath it does not. When the price finally re-rates, it re-rates up to the usage that was there the whole time. The work is to already own the thing before it does.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Open Source Switched Sides]]></title><description><![CDATA[Open source was the hedge against concentrated power. In 2026 its biggest sponsor is the Chinese state, and value moves accordingly.]]></description><link>https://blog.canonical.cc/p/open-source-switched-sides</link><guid isPermaLink="false">https://blog.canonical.cc/p/open-source-switched-sides</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Fri, 17 Jul 2026 15:37:17 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!bMOe!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Open source began as a hedge against concentrated power. Linux, Apache, Python, Kubernetes, RISC-V: foundational software built by people who did not want any single company owning the layer everyone else depends on. It was the closest thing computing had to a commons, and for thirty years it was mostly a Western, hobbyist, faintly libertarian project.</p><p>In 2026, the largest sponsor of free frontier AI is the Chinese state.</p><p>Start with the facts. The open models that keep landing at or near the frontier are increasingly Chinese. DeepSeek, Qwen, and Moonshot&#8217;s Kimi now trade blows with the best American models on real benchmarks, ship under permissive licenses, and arrive within months of the closed frontier. The old story was that China was a cheap fast-follower, a year back at a tenth of the price. That story is out of date. The gap is now measured in months, and the strongest Chinese labs are confident enough to charge close to full price for a hosted version while still giving the weights away.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!bMOe!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!bMOe!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 424w, https://substackcdn.com/image/fetch/$s_!bMOe!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 848w, https://substackcdn.com/image/fetch/$s_!bMOe!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 1272w, https://substackcdn.com/image/fetch/$s_!bMOe!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!bMOe!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png" width="1364" height="850" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/2a58f227-b979-456e-924f-228621c9b617_1364x850.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:850,&quot;width&quot;:1364,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:491112,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/207430541?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!bMOe!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 424w, https://substackcdn.com/image/fetch/$s_!bMOe!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 848w, https://substackcdn.com/image/fetch/$s_!bMOe!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 1272w, https://substackcdn.com/image/fetch/$s_!bMOe!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2a58f227-b979-456e-924f-228621c9b617_1364x850.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>This is deliberate commoditization, sanctioned from the top. If you cannot win the frontier outright, the next best outcome is to make sure the frontier is not worth much. Open weights turn the core asset of OpenAI and Anthropic into a free good. They also set the default. Every developer who builds on a Chinese open model is one more person fluent in China&#8217;s tools, standards, and assumptions. The United States understood this logic perfectly when it gave GPS away and let the world build on top of it. Whoever owns the standard owns the leverage that comes with it. China is running that playbook one layer up, at the model.</p><p>There is a sharp irony underneath this. Export controls were meant to protect an American compute lead. What they actually did was force Chinese labs to get radically better at doing more with less. Starved of the newest chips, they leaned into efficiency: sparser models that wake only a fraction of their parameters per token, cheaper attention, aggressive compression. The same thing happened in lithography, where a constrained foundry reached advanced nodes on older equipment because it had no other option. Constraint is a forcing function. The controls meant to widen the moat are training the discipline that drains it.</p><p>This is the part worth sitting with. Open source was supposed to be insurance against concentrated power. Today the biggest force pushing free frontier intelligence into the world is an authoritarian state. We still believe open beats closed. We wrote that recently and we mean it. What changed is not the direction, it is who benefits.</p><p>The United States has no clean answer. Banning Chinese open models mostly binds American firms: weights spread by torrent, so you get costlier domestic AI while the rest of the world standardizes on Chinese models anyway. Winning the open game directly needs a lab both willing to open a frontier model and able to build one, and no American lab is both. Either path cedes the default.</p><p>For us, the investing implication follows directly. If the model layer is commoditizing and going open, and it is, then durable value does not sit in the weights. It sits in what open weights cannot commoditize: verifiable inference, private deployment, the sovereign and enterprise builds that cannot run on a foreign API, and the distribution that turns a free model into something people actually pay for. Open winning does not mean nobody captures value. It means value moves, and it moves to exactly the layers we have been backing.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!d9LX!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!d9LX!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 424w, https://substackcdn.com/image/fetch/$s_!d9LX!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 848w, https://substackcdn.com/image/fetch/$s_!d9LX!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 1272w, https://substackcdn.com/image/fetch/$s_!d9LX!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!d9LX!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png" width="650" height="461.1607142857143" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1033,&quot;width&quot;:1456,&quot;resizeWidth&quot;:650,&quot;bytes&quot;:620343,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/207430541?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!d9LX!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 424w, https://substackcdn.com/image/fetch/$s_!d9LX!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 848w, https://substackcdn.com/image/fetch/$s_!d9LX!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 1272w, https://substackcdn.com/image/fetch/$s_!d9LX!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F879f1d61-6517-4a7b-bc28-c44d20185fcc_2194x1556.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The world&#8217;s default intelligence may end up open, cheap, and not American. That is not a prediction we love. It is one we would rather underwrite than ignore.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[The Transistor Was the Easy Part]]></title><description><![CDATA[The next decade in chips will be won on integration, not invention.]]></description><link>https://blog.canonical.cc/p/the-transistor-was-the-easy-part</link><guid isPermaLink="false">https://blog.canonical.cc/p/the-transistor-was-the-easy-part</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 10 Jul 2026 12:15:24 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/38388ddc-31ee-4fb0-83d2-d50d01cd9816_1088x608.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>For fifty years, a better chip meant one thing: a smaller transistor. Every roadmap, every fab, every $300M lithography machine pointed at the same lever. Shrink the feature, double the density, wait two years, repeat. That lever has largely stopped moving, and if you are not TSMC it stopped a while ago. The reflex is to read this as the end of something. We read it as the start of the most interesting decade in hardware since the integrated circuit itself.</p><p>The reason is a collision. Demand for compute is going vertical at the exact moment the one trick the whole industry relied on has run out of room. Power, not chips, is now the binding constraint on AI: this year alone <a href="https://www.tomshardware.com/tech-industry/artificial-intelligence/half-of-planned-us-data-center-builds-have-been-delayed-or-canceled-growth-limited-by-shortages-of-power-infrastructure-and-parts-from-china-the-ai-build-out-flips-the-breakers">billions of dollars of datacenter capacity has been delayed or cancelled</a> not for lack of funding or silicon, but because the grid cannot deliver electrons fast enough. And a large share of the energy we do deliver is spent not computing but shuttling data between memory and processor. But progress does not stop. It moves, and right now it is moving in three directions at once: new materials, new geometries, and new architectures.</p><p><strong>New materials.</strong> There is a class of device that holds memory and does the computation in one place, the way a synapse does, instead of separating the two. <a href="https://www.nature.com/articles/s41586-025-08733-5">The physics has been understood for roughly two decades</a> and the performance ceiling is enormous, but none of these devices are in production. The reason: the materials involved have historically been impossible to build into a normal silicon chip without destroying the transistors underneath.</p><p><strong>New geometries.</strong> Photonic chips do not need three-nanometer features. They need three-dimensional, multi-material, <a href="https://stateofthefuture.substack.com/p/the-asml-killer">wavelength-scale structures that the world&#8217;s most expensive machines were simply never built to make</a>. The unlock there is not smaller. It is a different shape entirely, and the tools optimized for shrinking transistors cannot produce it.</p><p><strong>New architectures.</strong> Once you can co-locate memory and compute, you get to attack the von Neumann bottleneck directly rather than papering over it with faster interconnects and more cache. The prize is doing the core operation of a neural network in a single physical step instead of thousands of read-write shuffles. That is where the order-of-magnitude efficiency gains live. And when power is the constraint, efficiency is the only thing that matters.</p><p>In every one of these cases, the breakthrough is not the invention. It is the integration. The exotic material, the photonic structure, and the analog crossbar have existed in labs for years. What&#8217;s been missing is a way to manufacture them at scale without throwing away the trillion-dollar silicon supply chain. The companies we think win this decade are those who figure out how to build these devices on machines the fabs already own, inside an existing process. Complement the incumbent, do not try to replace it. The moat moved from what you can invent to what you can integrate, and that is a much less crowded, much more defensible place to stand.</p><p>This reframes what we look for. In diligence, we care less about a benchmark from a clean-room prototype than whether the process survives contact with a real foundry, which is where almost everyone stalls. It also reopens the map. The leading-edge logic race is lost outside Taiwan and a single town in the Netherlands, but the new wedges are early, unoptimized, and open to new entrants in a way the transistor race has not been for thirty years. That is why the best teams we see are not trying to out-TSMC TSMC. They are building what TSMC was never designed to build.</p><p>Why now? AI created both halves of the opportunity. It created the demand by making energy efficiency existential. And it supplied the tools to meet it: cheap GPU compute, better lasers, a consolidated supply chain. These approaches turned viable in the last twenty-four months, not the last twenty years. The constraint and the cure arrived together.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[Why We Read Papers]]></title><description><![CDATA[Why our sourcing runs through the lab, not just the accelerator]]></description><link>https://blog.canonical.cc/p/why-we-read-papers</link><guid isPermaLink="false">https://blog.canonical.cc/p/why-we-read-papers</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Sat, 04 Jul 2026 02:48:55 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/fe814e54-50da-4b89-9d73-958d7ffff855_1088x608.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Some of the most valuable companies in the world, <a href="https://en.wikipedia.org/wiki/History_of_Google">Google</a> and <a href="https://en.wikipedia.org/wiki/Databricks">Databricks</a> among them, started as research projects in university labs, years before they had a name, a brand, a deck, or a fundraising round. They start as the work a small group does at the niche edge of a field most of the market is unaware of. That is where we spend a disproportionate share of our sourcing energy, and this week gave us a great proof point of that strategy.</p><p>Arena announced a <a href="https://x.com/arena/status/2071630464583151727">$100M annual revenue run rate</a>, eight months after launching its evaluation product. It started as a <a href="https://sky.cs.berkeley.edu/project/chatbot-arena/">research project at UC Berkeley</a>, with one mission: to measure AI progress through real-world use. The company is barely a year into commercialization, but that research began back in 2023. By the time the revenue showed up, the hardest part was already done: earning the whole field&#8217;s trust as the neutral scoreboard for AI.</p><p>That gap is what we keep coming back to. Arena&#8217;s team figured out how to measure model utility before it was even recognized as a central problem by industry. The research was the moat, and the revenue is the market catching up to a bet placed long ago in a lab. We would rather get to know these people while they are still publishing than compete for them once every fund has the same warm intro.</p><p>This is why our focus on academics is so deliberate. At the frontier, in AI, robotics, energy, and cryptography, the durable edge tends to be technical depth and being early to a paradigm. Academics live there by definition. They often see the shape of the next S-curve before it has a name, and they carry conviction that is hard to reverse-engineer from a competitor&#8217;s launch.</p><p>It also shapes where we look. We think the best technical founders are spread far wider than the target-school shortlist suggests, so we put as much energy into the research groups at UIUC, UT Austin, and Georgia Tech as we do at Stanford or MIT. The work coming out of the less-trafficked programs is world-class, and the founders behind it are often the least contested. Getting to know them early is one of the most underrated edges we have found in venture.</p><p>So we try to treat the lab as the top of the funnel. We map research groups, we read the papers, and we build relationships years before anyone incorporates. Arena is a good reminder of why that patience compounds. The company was there for anyone paying attention, long before it printed a nine-figure run rate. The best companies are often visible on paper first. Our job is simply to be reading.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item><item><title><![CDATA[The Custom Silicon Wave: Why AI Hardware Just Changed Forever]]></title><description><![CDATA[The era of generalist GPUs is giving way to hyper-specialized custom silicon. Here is what the sudden hardware wave means for tech moats, venture capital, and the cost of compute.]]></description><link>https://blog.canonical.cc/p/the-custom-silicon-wave-why-ai-hardware</link><guid isPermaLink="false">https://blog.canonical.cc/p/the-custom-silicon-wave-why-ai-hardware</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Thu, 02 Jul 2026 10:31:02 GMT</pubDate><content:encoded><![CDATA[<p>The AI hardware world just went through an incredibly busy two weeks. In late June, custom chips went from ideas on a page to working physical hardware. OpenAI, Etched, Amazon, and SambaNova all made massive moves at the exact same time.</p><p>To understand why, think of general graphics processors (GPUs) as world-class generalist chefs. They can cook any dish, but they spend a lot of time cleaning up, reading recipes, and finding ingredients. Custom chips (called ASICs) are like specialized sushi chefs who only slice tuna. They do just one job, but they do it incredibly fast with zero wasted effort.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>In late June, the industry decided it was time to bring in the sushi chefs.</p><h3>Inside the June Silicon Wave</h3><p>The velocity of this hardware wave is unprecedented. Four major players moved almost simultaneously:</p><ul><li><p><strong>OpenAI&#8217;s &#8220;Jalape&#241;o&#8221;:</strong> Partnering with Broadcom and Celestica, OpenAI built its first custom chip in just nine months. Early test samples are already running internal machine learning workloads in the lab. They are planning to deploy these custom chips at a massive <a href="https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration/">10-gigawatt scale by 2029</a>.</p></li><li><p><strong>Etched&#8217;s &#8220;Sohu&#8221;:</strong> This startup came out of stealth with a TSMC 4nm chip designed <em>purely for transformer models</em>. They skipped all the usual general-purpose parts to focus only on AI math. They reached first-pass silicon success, raised $800 million, and booked over $1 billion in forward contracts. <a href="https://www.etched.com/">Etched claims</a> their 8-chip server can run Meta&#8217;s Llama 70B model at 500,000 tokens/s! (while this is a self-reported claim that has not been independently verified yet) This shows how fast custom chips can be.</p></li><li><p><strong>Amazon&#8217;s Trainium:</strong> For over 10 years, Amazon kept its custom chips strictly inside its own cloud. Now, they are in early talks <a href="https://www.benzinga.com/markets/tech/26/06/53285203/amazon-expands-trainium-ai-chip-strategy-beyond-aws">to sell their directly to other data centers</a>. This is a massive shift that challenges Nvidia&#8217;s crown.</p></li><li><p><strong>SambaNova&#8217;s SN50:</strong> This startup launched its fifth-generation AI chip aimed at making enterprise AI cheaper and easier to run. Their system can run in standard, air-cooled data centers so companies do not have to rebuild their facilities. SambaNova is looking to raise up to $1 billion at a <a href="https://www.nextplatform.com/ai/2026/02/25/sambanova-pits-its-engineering-against-nvidia-for-agentic-ai/4092613">$10 billion valuation</a>.</p></li></ul><h3>Rethinking the &#8220;Moat&#8221;</h3><p>This wave of custom hardware has venture capital rethinking what makes an AI company defensible. For years, investors poured money into software startups that optimize models to run on generic GPUs. But as giant tech companies build custom chips that hardwire these optimizations directly into the silicon, those software-only moats are getting squeezed.</p><p>But this does not mean the infrastructure layer is dead. As <a href="https://tomtunguz.com/what-if-there-is-no-moat/">Tunguz said</a> moats do not always have to be there on day one. While hardware startups usually need a massive technical lead to start, enduring moats can actually be earned over time through relentless execution, branding, and distribution. He points for example to Salesforce, which won the cloud CRM market over technically superior rivals simply because they executed and scaled better over ten years.</p><h3>The New Bottom Line: Hardware Dictates the Game</h3><p>The AI race is no longer just about who can write the best algorithms. Software is quickly becoming a cheap commodity. The real, defining advantages of the next decade are <strong>physical</strong>: designing custom silicon, securing power grids, and building efficient hardware. The players who win the physical layer will control the future of AI.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Power In, Heat Out: An AI Data Center Primer]]></title><description><![CDATA[Why power, not chips, is the real bottleneck. And why AI is both the strain on the buildout and the tool finishing it.]]></description><link>https://blog.canonical.cc/p/power-in-heat-out-an-ai-data-center</link><guid isPermaLink="false">https://blog.canonical.cc/p/power-in-heat-out-an-ai-data-center</guid><dc:creator><![CDATA[Canonical]]></dc:creator><pubDate>Tue, 23 Jun 2026 13:03:21 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!buDX!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<h2>Red dirt in Johor</h2><p>Last week a couple of days after my panel at <a href="https://www.superai.com/?ref=canonicalcc">SuperAI</a> in Singapore on &#8220;<a href="https://www.youtube.com/watch?v=WKW2T1oQTME&amp;ref=canonicalcc">Compute, Chips and the Cost of Intelligence</a>&#8220;, I stood on a cleared lot in Johor, Malaysia, watching a future AI data center get staked out. Everyone on that panel was talking about GPUs. But what decides whether that lot in Johor becomes a working cluster is not just chips. It is whether you can get a gigawatt of power to the dirt, cool it, and find people who know how to run it. That gap is what this post is about.</p><p>It is a companion to our map at <a href="https://www.canonical.cc/labs/data-centers/">canonical.cc/labs/data-centers</a>, which tracks 31 players across 7 layers. It sits next to our <a href="https://www.canonical.cc/labs/semiconductor-silicon-stack/">Semiconductor Stack</a> and <a href="https://www.canonical.cc/labs/decentralized-ai/">Decentralized AI</a> maps. Body, brain, nervous system of the buildout.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><h2>The strange loop</h2><p>Our silicon primer was predicated on this fact: </p><p><em><strong>the technology breaking the chip industry is the same technology most likely to fix it.</strong></em> </p><p>AI demand is breaking the grid. A single campus now draws a gigawatt, a small city&#8217;s worth of power for one building. And AI is turning out to be the best tool we have for designing those power systems, tuning the cooling, and squeezing more compute from concrete we already poured.</p><h2>A building, on a napkin</h2><p>A data center is a building full of computers. Power comes in, heat comes out, compute happens in between. That is the whole thing. We went from server rooms, to the cloud, to the AI explosion, and the physics never changed. Only the scale did.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!buDX!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!buDX!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 424w, https://substackcdn.com/image/fetch/$s_!buDX!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 848w, https://substackcdn.com/image/fetch/$s_!buDX!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 1272w, https://substackcdn.com/image/fetch/$s_!buDX!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!buDX!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png" width="1456" height="760" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/ecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:760,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:214022,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/203062234?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!buDX!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 424w, https://substackcdn.com/image/fetch/$s_!buDX!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 848w, https://substackcdn.com/image/fetch/$s_!buDX!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 1272w, https://substackcdn.com/image/fetch/$s_!buDX!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd04734-2e10-41a7-a53e-bdf0a480462c_1732x904.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The number everyone quotes is PUE, Power Usage Effectiveness: total power divided by the power that reaches the chips. A PUE of 1.2 means you burn 0.2 watts on cooling for every watt of compute. The old server room ran near 2.0. Liquid-cooled AI halls are pushing toward 1.1. At gigawatt scale, that gap is a power plant&#8217;s worth of waste.</p><p>Now stack it up. From the rental product at the top to the grid at the bottom, every layer is its own market.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!pbZl!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!pbZl!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 424w, https://substackcdn.com/image/fetch/$s_!pbZl!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 848w, https://substackcdn.com/image/fetch/$s_!pbZl!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 1272w, https://substackcdn.com/image/fetch/$s_!pbZl!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!pbZl!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png" width="1400" height="1372" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1372,&quot;width&quot;:1400,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:296532,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/203062234?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!pbZl!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 424w, https://substackcdn.com/image/fetch/$s_!pbZl!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 848w, https://substackcdn.com/image/fetch/$s_!pbZl!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 1272w, https://substackcdn.com/image/fetch/$s_!pbZl!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F73bc761f-322b-448d-b40e-cdddcd3a77ed_1400x1372.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><strong>1. Neoclouds.</strong> The product most AI companies buy: GPUs by the hour, no enterprise baggage. <a href="https://www.coreweave.com/">CoreWeave</a>, <a href="https://www.crusoe.ai/">Crusoe</a>, <a href="https://lambda.ai/">Lambda</a>, <a href="https://nebius.com/">Nebius</a>. CoreWeave proved the category reaches public scale. It also exposed the category&#8217;s central risk, below.</p><p><strong>2. Operations.</strong> Who runs the cluster once it is powered on. The quietest layer on the map, and we think the most mispriced. More below.</p><p><strong>3. Compute silicon.</strong> The GPUs and accelerators. <a href="https://www.nvidia.com/">NVIDIA</a>, <a href="https://www.amd.com/">AMD</a>, Google TPU, <a href="https://www.cerebras.ai/">Cerebras</a>. Everyone fights here. We cover it in the silicon map, so we will skip it.</p><p><strong>4. Networking.</strong> Past 100,000 GPUs, the bottleneck stops being compute and becomes moving data between chips. <a href="https://www.arista.com/">Arista</a>, <a href="https://www.broadcom.com/">Broadcom</a>, optical-I/O upstarts like <a href="https://ayarlabs.com/">Ayar Labs</a>.</p><p><strong>5. Cooling.</strong> Old racks drew 5 to 20 kilowatts; air handled it. AI racks draw 40 to 80 today, and NVIDIA is targeting a megawatt per rack on Rubin Ultra. Air cannot touch that. Liquid goes straight to the chip. <a href="https://www.motivaircorp.com/">Motivair</a>, <a href="https://www.coolitsystems.com/">CoolIT</a>, <a href="https://jetcool.com/">JetCool</a>.</p><p><strong>6. Power.</strong> The binding constraint. <a href="https://www.constellationenergy.com/">Constellation</a> restarting Three Mile Island, <a href="https://oklo.com/">Oklo</a> and small reactors, plus a fast-growing layer of on-site generation and batteries that smooth a training run&#8217;s spiky draw.</p><p><strong>7. Facilities.</strong> The buildings and land, mostly held in REITs. <a href="https://www.equinix.com/">Equinix</a>, <a href="https://www.digitalrealty.com/">Digital Realty</a>, <a href="https://vantage-dc.com/">Vantage</a>, <a href="https://www.qtsdatacenters.com/">QTS</a>.</p><h2>So where does it break?</h2><p>Which layer can actually stop the buildout?</p><p>In silicon, the answer was two companies: ASML and TSMC. Here it is not a company. It is power.</p><p>About <a href="https://emp.lbl.gov/queues?ref=canonicalcc">2,300 gigawatts of generation sit stuck in US interconnection queues</a>, more than the country&#8217;s entire installed capacity. </p><p>The average project waits five years to connect. <a href="https://ifp.org/interconnection-for-ai/?ref=canonicalcc">Some data centers are quoted twelve</a>. </p><p>US data-center demand is headed for <a href="https://www.spglobal.com/energy/en/news-research/latest-news/electric-power/101425-data-center-grid-power-demand-to-rise-22-in-2025-nearly-triple-by-2030?ref=canonicalcc">roughly 76 gigawatts in 2026</a>, up from 50 in 2024. </p><p>In Texas, <a href="https://www.datacenterdynamics.com/en/news/centerpoint-energy-reports-700-percent-increase-in-data-center-interconnection-requests-in-texas/?ref=canonicalcc">large-load requests to one utility jumped sevenfold</a> in a year. </p><p><a href="https://www.bvp.com/atlas/roadmap-the-ai-data-center-stack?ref=canonicalcc">Bessemer counts 190 gigawatts</a> of hyperscale capacity already announced against 5-7 year queues.</p><p>&#8220;We need more compute&#8221; is something we are all hinding behind. The world is not just short on chips this quarter. It is short on energized, cooled, staffed megawatts. You can panic-order GPUs. You cannot panic-build a substation, and you cannot panic-train the people who run the room.</p><h2>Where AI causes the squeeze</h2><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!0vo-!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!0vo-!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 424w, https://substackcdn.com/image/fetch/$s_!0vo-!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 848w, https://substackcdn.com/image/fetch/$s_!0vo-!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 1272w, https://substackcdn.com/image/fetch/$s_!0vo-!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!0vo-!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png" width="1456" height="1023" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1023,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:301268,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://blog.canonical.cc/i/203062234?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!0vo-!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 424w, https://substackcdn.com/image/fetch/$s_!0vo-!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 848w, https://substackcdn.com/image/fetch/$s_!0vo-!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 1272w, https://substackcdn.com/image/fetch/$s_!0vo-!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7b249088-c974-4158-b3bf-a29f364a3813_1844x1296.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Power takes the direct hit. Cooling takes the next one, as rack density climbs an order of magnitude and liquid goes from exotic to mandatory.</p><p>Then the strangest one: utilization. The most expensive hardware ever built mostly sits idle. xAI&#8217;s fleet <a href="https://wccftech.com/xai-using-just-11-percent-gpus-while-meta-google-squeeze-out-much-more/?ref=canonicalcc">reportedly runs near 11%</a>. <a href="https://www.fool.com/earnings/call-transcripts/2026/05/07/cloudflare-net-q1-2026-earnings-call-transcript/?ref=canonicalcc">Hyperscalers run 5-10%</a>. We spent 25 years learning to share CPUs with VMs, containers, and schedulers. Almost none of that exists for GPUs yet. So we scream about a shortage while most of our GPUs sit parked.</p><p>Under all of it, people. Maybe a few hundred humans on earth have run a multi-thousand-GPU cluster end to end. The demand is the entire buildout. Construction is no better: the trade is <a href="https://itif.org/publications/2026/01/12/construction-industry-facing-worker-shortage-driven-by-growth-of-data-centers/?ref=canonicalcc">short hundreds of thousands of workers</a>, the <a href="https://uptimeinstitute.com/resources/research-and-reports/uptime-institute-global-data-center-survey-results-2025?ref=canonicalcc">average data-center worker is 53, and most operators cannot fill ops roles</a>. The pain lands where the talent is thinnest. Not a coincidence.</p><h2>Where AI relieves it</h2><p>The same intelligence stressing the grid is being pointed back at the stack. AI permitting software is compressing the slowest, most bureaucratic step in the build (<a href="https://www.paces.com/">Paces</a>, Lorica). AI runs cooling and power inside live halls (<a href="https://www.phaidra.ai/">Phaidra</a>, <a href="https://www.corintis.com/">Corintis</a>). A new layer makes GPU work power-flexible, so a cluster leans into cheap power and backs off when the grid is tight (<a href="https://www.emeraldai.co/">Emerald AI</a>, Verse). Construction is getting robots (<a href="https://www.geckorobotics.com/">Gecko Robotics</a>, DroneDeploy).</p><p>The highest-leverage bet sounds the dullest. Build &#8220;VMware for GPUs,&#8221; real multi-tenancy and scheduling, and a 10 percent utilization problem becomes a 10x supply unlock. That is more new compute than most fabs add in years, and it ships in software.</p><p>Will any of this beat the power constraint this decade? No. Electrons obey physics and permitting, not roadmaps. But the software-shaped layers (permitting, orchestration, utilization, operations) compound in quarters, not half-decades. That is where venture lives.</p><h2>Sovereign AI: the repatriation trend worth watching</h2><p>For two decades the logic was centralize: ship compute to whoever had the cheapest power. That is reversing, fast. Countries now treat AI compute as critical infrastructure, like a grid or a port, and they want it on home soil under home law. Some call it geopatriation. I call it the most important capital-flows story in infrastructure right now.</p><p><span data-color="rgb(26, 26, 26)" style="color: rgb(26, 26, 26);">The EU has put </span><a href="https://ec.europa.eu/commission/presscorner/detail/en/ip_25_467?ref=canonicalcc">20 billion euros behind AI Gigafactories</a><span data-color="rgb(26, 26, 26)" style="color: rgb(26, 26, 26);">. France is building a </span><a href="https://www.france24.com/en/europe/20250207-uae-to-invest-up-to-%E2%82%AC50-billion-in-massive-ai-data-centre-in-france?ref=canonicalcc">1-gigawatt campus with the UAE</a><span data-color="rgb(26, 26, 26)" style="color: rgb(26, 26, 26);"> worth tens of billions, while backing </span><a href="https://mistral.ai/?ref=canonicalcc">Mistral</a><span data-color="rgb(26, 26, 26)" style="color: rgb(26, 26, 26);"> as its champion. Saudi Arabia stood up </span><a href="https://www.pif.gov.sa/en/news-and-insights/press-releases/2025/hrh-crown-prince-launches-humain-as-global-ai-powerhouse/?ref=canonicalcc">HUMAIN</a><span data-color="rgb(26, 26, 26)" style="color: rgb(26, 26, 26);"> under its sovereign fund to build the whole stack. The UAE is </span><a href="https://www.semafor.com/article/04/13/2026/data-centers-under-fire-test-gulf-sovereign-ai-ambitions?ref=canonicalcc">breaking ground on a multi-gigawatt cluster in Abu Dhabi</a><span data-color="rgb(26, 26, 26)" style="color: rgb(26, 26, 26);">. Japan and Singapore are moving the same way. We see it at seed stage too, with founders pitching energy-first inference backbones built for Europe.</span></p><p>Repatriation does not shrink the stack, it widens it. Every sovereign build needs its own power, its own cooling, and its own operators, in a country that has never run a frontier cluster. The talent gap that is acute in Virginia becomes a wall in Riyadh or Johor. Many national buildouts means the power and operations layers get demanded everywhere at once. That is the picks-and-shovels case, multiplied by the flags on the map.</p><h2>Why this could fail</h2><p>I would be selling you a dream if I skipped the bear case. It is real and specific.</p><p>The sharpest risk is financial. Neoclouds buy GPUs with GPU-backed debt. If utilization or pricing softens, the collateral can depreciate faster than the loans amortize, and one demand wobble cascades through the most leveraged layer. CoreWeave&#8217;s bull case and bear case are the same fact. Add regulatory backlash (moratoriums, water fights, grid-reliability fights are already spreading) and the chance that hyperscalers re-bundle and fix the allocation problems that created the neocloud layer in the first place. And the deepest question stays open: does inference get effectively free as efficiency compounds, or does demand outrun supply forever, keeping power the permanent constraint? The thesis hinges on which way that breaks.</p><p>Our read: the bull case is already priced into the obvious layers, silicon and the public neoclouds. We would rather underwrite where the centralized option is structurally weak and demand is non-negotiable. Power. Liquid cooling. And the operators who turn a half-billion-dollar building full of idle silicon into a working cluster.</p><h2>The exciting part</h2><p>Today, standing up AI compute takes a hyperscaler&#8217;s balance sheet and a team that barely exists. Productize the operations and power layers and that capability gets unbundled: handed to the sovereign, the enterprise, the neocloud in a country you would not have guessed. Standing up a cluster starts to look less like building a refinery and more like provisioning a service.</p><p>Somewhere on that lot in Johor, rebar is going in for a building that will think. Whether it ever does comes down to the least glamorous layers on the map: the wire coming in, the heat going out, and the few people who can keep it alive at 3am. That is the layer we are building toward.</p><p>The flashy layers (silicon, the public neoclouds) are mostly priced. The middle (power, cooling, operations) is where we think the next infrastructure companies get built, and where AI is the relief, not the strain. We track all of it, layer by layer, with sources and caveats, on the full map.</p><p><strong>Explore the map: <a href="https://www.canonical.cc/labs/data-centers/">Data Centers &amp; Neoclouds at Canonical Labs &#8594;</a></strong></p><p><em>Educational, not investment advice. Figures are point-in-time as of June 2026.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[One Model Won't Win]]></title><description><![CDATA[Open models are near-frontier and nearly free. The advantage just moved off the model.]]></description><link>https://blog.canonical.cc/p/one-model-wont-win</link><guid isPermaLink="false">https://blog.canonical.cc/p/one-model-wont-win</guid><dc:creator><![CDATA[Anthony Avedissian]]></dc:creator><pubDate>Fri, 19 Jun 2026 17:48:36 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/ffe76310-cd0c-4d47-a9c5-0566f435d4a4_1088x608.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>For the first time, picking a model is no longer an edge. Open-weight models now land within a few points of the frontier on most benchmarks while <a href="https://inference.net/content/llm-api-pricing-comparison/">costing 50 to 90% less to run</a>, and API prices have fallen more than 90% since 2023. When near-frontier intelligence is available to everyone for pennies, the model stops being the moat. It becomes a commodity input.</p><p>When the input commoditizes, value moves to whoever orchestrates it. That is happening in two directions at once, above the model and below it.</p><p>Above the model, the aggregator is becoming the intelligence layer. OpenRouter spent two years routing each request to the single best model. This month it launched <a href="https://x.com/OpenRouter/status/2065856853989270011">Fusion</a>, which fans a prompt out to a panel of models, has a judge reconcile their answers, and returns one synthesized result. The panel beats any single model: 69% on the DRACO research benchmark against 65% for the best solo model, and ahead of both GPT-5.5 and Opus 4.8. The routing layer stopped picking the winner and started manufacturing one.</p><p>Below the model, serving is becoming its own discipline. A cheap open model is only cheap if you serve it efficiently, and the optimization surface (batching, KV cache, speculative decoding) moves faster than any one team can track. <a href="https://github.com/vllm-project/vllm">vLLM</a> is the floor, not the ceiling, and the frontier shifts week to week. The edge is no longer which model you run. It is how few GPUs you need to run it.</p><p>We think this is where the margin goes. Not to whoever trains the best model, but to whoever orchestrates models best on top and serves them cheapest underneath. We&#8217;ve said before that open weights commoditize the base layer and the moat moves up-stack. The model was the product for three years. Now it is the raw material, and the companies that matter are the ones that turn it into something cheaper, faster, or smarter than any single model could be alone.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://blog.canonical.cc/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading!</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p></p>]]></content:encoded></item></channel></rss>