{"id":"2092981626716508495","url":"https://x.com/ttunguz/status/2092981626716508495","text":"From −94% gross margins to 0m of revenue per megawatt: how AI models learned to print money. https://x.com/i/article/2092981370255548416","author":{"name":"Tomasz Tunguz","username":"ttunguz","avatarUrl":"https://pbs.twimg.com/profile_images/901542400559992832/yDp0b2Al_200x200.jpg"},"createdAt":"Thu Aug 27 14:24:40 +0000 2026","engagement":{"replies":8,"retweets":15,"likes":138,"views":21248},"article":{"title":"Revenue per Megawatt & The AI Model Factory","previewText":"An AI model company buys wholesale electricity by the megawatt & resells it as cognitive work.\n“The base cost of compute tends to be around 10 or 13 or $15 million per megawatt. In the case of","coverImageUrl":"https://pbs.twimg.com/media/HQvDh43bYAAhpe2.jpg","content":"An AI model company buys wholesale electricity by the megawatt & resells it as cognitive work.\n\n> “The base cost of compute tends to be around 10 or 13 or $15 million per megawatt. In the case of Anthropic, the revenue has gone as high as $50 million per megawatt. And what that now enables them to do is, hey, if I spend 10 bucks on inference capacity, I actually generate 50 bucks of revenue. And then I can turn around and incrementally spend all of that profit on training.”— Dylan Patel, SemiAnalysis, on the Dwarkesh Podcast[1](http://localhost:1313/revenue-per-megawatt/#fn:1)\n\nGross profit measured as a function of electricity proves model companies can be profitable on a contribution basis.\n\nAnthropic’s gross margin was −94% in 2024 : $1.94 of compute for every $1 of revenue.[2](http://localhost:1313/revenue-per-megawatt/#fn:2)\n\n![](https://pbs.twimg.com/media/HQvDkByaEAAa9eK.png)\n\nBy 2025 the corner turned, & Anthropic swung from −94% to a 40-50% gross margin.\n\nIn 2026 revenue passed cost : $50m per megawatt against a $10-15m cost. Anthropic booked its first profitable quarter, $10.9b of revenue & $559m of operating profit.\n\nThat 5% operating margin sits well below the 70 to 80% gross margin the megawatt math implies. Training runs & headcount consume the difference.\n\n![](https://pbs.twimg.com/media/HQvDmmSa8AAPmGA.jpg)\n\nGross profit per megawatt is not just about intelligence, but also efficiency. A model that serves the same intelligence at a fraction of the compute generates more profit per megawatt, even at a lower price.\n\nGLM-5.3-Flash scores 57 on the Artificial Analysis Intelligence Index : identical to Claude Opus 4.8.[3](http://localhost:1313/revenue-per-megawatt/#fn:3) But it does it on 18 billion active parameters, at a 90 to 97% reduction in cost. Fewer active parameters & less attention compute means more tokens per megawatt.\n\n![](https://pbs.twimg.com/media/HQvDpWSb0AEC_WL.png)\n\nThe frontier itself keeps climbing in waves. Three jumps of 3 points or more carry half of the 25-point gain from 37 to 63. The efficient models chase a target that resets every quarter.\n\nWhich is what the profit per megawatt buys. Patel’s last clause is the hinge : “turn around and incrementally spend all of that profit on training.” The margin funds the factory.\n\nNvidia paid $6b to acquire Poolside & invested another $1b on that premise. Model building becomes an industrial process : thousands of experiments across a search space, not artisanal hand-tuning.\n\n> “The model is the output. The ability to keep building better models, faster & more efficiently each time is the actual innovation. The Model Factory is the compounding asset.”— Jason Warner, Poolside[4](http://localhost:1313/revenue-per-megawatt/#fn:4)\n\nLaguna S 2.1 went from kickoff to release in 52 days.\n\nInference margin funds the factory that makes the next model cheaper to build & more efficient to run.\n\n1. Dylan Patel on the Dwarkesh Podcast, “Anthropic & OpenAI will have most of the world’s compute by 2028” (Aug 2026). [Apple Podcasts](https://podcasts.apple.com/us/podcast/dylan-patel-anthropic-openai-will-have-most-of/id1516093381?i=1000785793715) [↩︎](http://localhost:1313/revenue-per-megawatt/#fnref:1)\n\n1. The Information, “Anthropic’s Gross Margin Flags Long-Term AI Profit Questions.” [The Information](https://www.theinformation.com/articles/anthropics-gross-margin-flags-long-term-ai-profit-questions) [↩︎](http://localhost:1313/revenue-per-megawatt/#fnref:2)\n\n1. Artificial Analysis Intelligence Index. [Artificial Analysis](https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index) [↩︎](http://localhost:1313/revenue-per-megawatt/#fnref:3)\n\n1. Jason Warner, Poolside. [X](https://x.com/jasoncwarner/status/2092661422631330227) [↩︎](http://localhost:1313/revenue-per-megawatt/#fnref:4)"}}