Anthropic is in funding talks that would value the company near $2 trillion, according to Ars Technica's August reporting, at the same moment Ars reports OpenAI and Anthropic are cutting API prices as Chinese labs close the capability gap. The two stories are one story: the valuation is a bet on locked-in enterprise deployment, the API business banks and hospitals actually wired into their compliance workflows, not on which lab ships the best model this month. That is why the price cuts matter more than the valuation headline. When a Chinese lab like DeepSeek or Moonshot's Kimi ships a model that scores close to Claude or GPT-5 on public benchmarks at a fraction of the inference cost, the fastest lever a US lab has to defend its installed base isn't a better model, it's a lower price, because switching costs (the audit, the compliance sign-off, the systems integration) protect the contract but not forever if the price gap gets too wide.

The number that will decide whether $2 trillion is a real price or a talking point is the training compute bill behind it. Anthropic runs primarily on Amazon's Trainium 2 clusters and Google TPU capacity, not the Nvidia B200 NVL72 systems that OpenAI and xAI depend on, and that dependency is now colliding with Nvidia's own reported $500 billion capacity bet on aging Hopper and Blackwell-generation supply. A valuation built on enterprise contracts still has to clear a compute bill set by whichever chip generation its cloud partners commit next, and Amazon and Google's Q3 2026 capital expenditure guidance, due in their October earnings calls, is the next number that tells you whether that bill is one Anthropic's revenue can actually cover.