Anthropic Hit 80x Growth. Its Real Problem Was the Power Grid.
Anthropic planned for 10x growth in 2026. It hit 80x.
Dario Amodei, Anthropic's CEO, said that gap — the distance between the company's internal target and the actual demand pouring through its systems — is the reason the company has spent the past several months unable to serve all its customers fast enough. He made the admission at Anthropic's Code with Claude 2026 developer conference in San Francisco on May 6, according to CNBC, which first reported the remarks.
The numbers are stark. Run-rate revenue has now surpassed $30 billion, Anthropic confirmed at the same conference, up from approximately $9 billion at the end of 2025. API volume on the Anthropic platform is up 17x year-on-year. Amodei called the growth rate "just crazy" and "too hard to handle", and said he hopes for "more normal" expansion ahead.
The concrete result: Anthropic has run into the same constraint that every major AI lab is now racing to solve. The bottleneck is not the model. It is the power grid.
"The fact that serious companies are even discussing compute capacity in space tells you how aggressively the market is searching for power and scale," Ryan Mallory, CEO of Flexential, a data center operator, told Reuters last week. That comment was about the industry broadly. It applies to Anthropic specifically.
The company is mid-buildout. Reuters reported Anthropic has struck a deal to rent all 220,000 chips at SpaceX's Colossus 1 facility, giving it access to 300 megawatts of power dedicated to its systems. It has also committed more than $100 billion to Amazon Web Services technologies over ten years, securing up to 5 gigawatts of new capacity. The company declined to specify which graphics processors power its systems, citing competitive sensitivity.
The strain is visible in customer-facing decisions. Anthropic doubled Claude Code five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, and removed peak-hours limit reductions for Pro and Max accounts — changes that function as a quiet acknowledgment that customers are hitting walls.
Amodei's most direct statement on the compute problem, per CNBC: "That is the reason we have had difficulties with compute."
One concrete signal of enterprise demand: Mercado Libre, the Latin American e-commerce and fintech company with 23,000 engineers, is aiming for 90% autonomous coding across its engineering workforce by Q3 2026, according to Amodei's remarks at the conference. If that target holds, it would be one of the most aggressive enterprise-wide AI coding deployments publicly disclosed by any company of that scale.
The irony is not lost on the industry. Anthropic built its reputation on the quality of its models. Its current crisis is not about benchmarks. It is about watts.
The forward contracts and data center deals Anthropic has signed are real. The gap between what it promised itself it could deliver and what customers actually consumed is also real — and it arrived faster than the infrastructure to support it. What comes next depends less on which lab ships the best model than on which ones can secure enough electricity to run them.
What to watch: whether Anthropic's infrastructure deals translate into reliable capacity before the next growth surge hits, and whether the rate-limit removals are a temporary fix or the start of a durable scaling response. Amodei's "more normal" growth target, if it holds, would itself be a significant deceleration from the 80x pace.