Exclusive: OpenAI Highlights Massive Compute Advantage Over Anthropic to Investors
It increasingly looks like Sam Altman was right and Anthropic is likely to pay a cost for underestimating the growth of AI compute demand.
Anthropic CEO Dario Amodei loves to portray himself as the responsible player, while suggesting his chief rival is a hyperaggressive, overspending drunken sailor.
At The New York Times DealBook conference in December, Dario said: “There are some players who are YOLO-ing. Who pull the risk dial too far. And I’m very concerned,” obviously implying it was OpenAI’s Sam Altman.
Dario reiterated the YOLO narrative on the Dwarkesh podcast in February, saying Anthropic is behaving responsibly: “I think it is true we’re spending somewhat less than some of the other players,” he said. “I get the impression that some of the other companies have not written down the spreadsheet, that they don’t really understand the risks they’re taking. They’re just doing stuff because it sounds cool.”
The media ate the narrative up, but it may be entirely wrong.
It increasingly looks like Sam was right and Anthropic is likely to pay a cost for underestimating the growth of AI compute demand.
OpenAI now knows it will benefit from its bigger bet on compute capacity.
Key Context reviewed a memo sent to OpenAI’s investors Wednesday with the subject line: “Compute is the ballgame.”
OpenAI said: “scale is now a real differentiator,” adding they have already identified over 8 gigawatts of capacity and “are planning 30GW by 2030.” According to a source familiar, OpenAI has line of sight on another 14GW of capacity (on top of the 8GW) toward the 30GW target by 2030.
OpenAI directly compares its own compute capacity ramp going from 0.2GW in 2023 to 0.6GW in 2024 to 1.9GW in 2025 versus its estimates for Anthropic’s ramp of 1.4GW for 2025, 3 to 4GW by end of 2026, and 7 to 8GW by end of 2027.
“Anthropic is operating on a meaningfully smaller curve .. Even at the high end of that range, our ramp is materially ahead and widening,” OpenAI wrote. “That gap matters because compute is now a product constraint.”
The memo argues compute capacity limits will eventually impede the ability to innovate on product and serve customers. I can concur, as when I use Claude it often goes down these days and becomes unusable due to too much demand.
OpenAI also correctly makes the argument that having more compute gives the company the ability to serve the mainstream public, who otherwise wouldn’t have access to the technology. Anthropic primarily serves users and enterprises through higher cost paid subscriptions.
“This leverage also enables OpenAI to continue democratizing AI. Instead of passing the costs of short-sightedness on compute to customers, we are able to make our tools available to hundreds of millions of people. It allows us to be more generous with builders, better partners to the developer ecosystem, and pass capacity on to the people who are using out tools to create and solve problems for themselves and society.”
There are clear signs the exponential demand for AI compute is accelerating. The industry has moved from basic chatbots to reasoning models and now to autonomous agents. Data from OpenRouter, which can be used as a proxy for AI token use, shows aggregate AI token demand is up 15 times year over year.
Over the last few months, the “killer app” driving this agentic surge is coding assistants. The low-latency speed and power required for these enterprise AI workloads is commanding premium pricing, driving surging revenue.
There is a gold rush here. On Monday, Anthropic revealed that its run-rate revenue has more than doubled in less than two months. As the company stated, “Our run-rate revenue has surpassed $30 billion, up from $9 billion at the end of 2025, as demand for Claude continues to accelerate.”
OpenAI has been pivoting resources toward enterprise and AI coding agents too. “Three million people are now using Codex weekly - up from two million a little under a month ago,” an OpenAI executive said earlier this week.
The surge is likely to continue. At GTC last month, Nvidia Chief Scientist Bill Dally and Google DeepMind Chief Scientist Jeff Dean shared a sense of urgency about the near-term explosion of agentic AI demand and the need to significantly improve hardware for faster and more capable inference. Jensen Huang has repeatedly said the “inference inflection” has arrived.
With AI compute undergoing liftoff as demand for agentic AI explodes, Anthropic is facing a problem with its more conservative growth assumptions. Compute is oxygen and Anthropic could run out of capacity to serve its customers amid rising demand.
OpenAI holds a massive advantage moving forward as more compute capacity may mean it can generate much more revenue than its main rival.
Sometimes betting big is the right move. The bold bet on capacity is looking prescient and OpenAI is set to capitalize.