Anthropic told staff its annualized revenue has reached $65 billion, up from roughly $1 billion two years ago, according to TechCrunch's August 17 report. Annualized revenue means taking the most recent month or quarter of income and multiplying it out to a full year, so the number moves fast in either direction, but the direction here has been almost straight up since Claude's enterprise API business started compounding contract renewals into training budget. The company is now in talks that would value it near $350 billion in its next funding round, and Ars Technica reported the same day that a public listing could value Anthropic near $2 trillion. Anthropic's compute for training runs comes overwhelmingly from Amazon's Trainium chips and AWS data centers under the companies' multi-year cloud agreement, the same dependency that lets a bank or hospital sign a Claude enterprise contract and trust the vendor has the hardware lined up for the next two years of model upgrades, not just the current one.

The number that should worry Anthropic's two US rivals more than the valuation is the one buried in the same week's Ars Technica report on the price war: Chinese labs are closing the gap on price just as Anthropic's revenue proves enterprise customers will pay a premium for reliability, not raw benchmark scores. That is the bet Anthropic's $2 trillion IPO math rests on, that a compliance officer at a bank who has already wired Claude into risk-modeling workflows will not rip it out for a cheaper Chinese model even if the eval scores converge, because the contract, the audit trail, and the AWS uptime guarantee do not come with the cheaper copy. Suzhou and Hsinchu-based manufacturers building the server racks for both sides of that price war do not care who wins it. Anthropic's IPO timeline, whenever it lands, is the first real market test of whether that enterprise stickiness is worth the multiple.