Anthropic considers its own chips
Sources say Anthropic is weighing designing its own AI chips as demand for Claude surges and cloud capacity tightens, signalling a push to reduce dependence on third-party hardware. The company is also using CoreWeave capacity to power Claude even as its reported run‑rate and liquidity moves attract regulatory scrutiny and questions about cyber risk from banks. (reuters.com) (mercurynews.com) (theguardian.com)
Anthropic is considering building its own artificial intelligence chips instead of relying entirely on outside suppliers, according to Reuters on April 9, and that is happening while the company scrambles for enough computing power to keep Claude running. (reuters.com) (cnbc.com) The immediate problem is simple: training and serving an artificial intelligence model requires huge numbers of specialized chips, and there still are not enough of them. Reuters reported Anthropic’s chip plans are early and could still end with the company sticking to bought-in hardware. (reuters.com) At the same time, Anthropic said on April 6 that Claude demand has accelerated so fast that its annualized revenue run rate passed $30 billion, up from about $9 billion at the end of 2025. The company also said more than 1,000 business customers are now spending over $1 million a year on an annualized basis. (anthropic.com) That surge helps explain why Anthropic is stacking up supply wherever it can find it. On April 10, CoreWeave announced a multi-year deal to provide computing capacity for the development and deployment of Claude, with the new capacity coming online later in 2026. (coreweave.com) (reuters.com) CoreWeave is not a household-name cloud company like Amazon Web Services or Google Cloud. It is a specialist that rents out graphics processing unit servers built for artificial intelligence workloads, which is why fast-growing model companies go there when the big clouds are tight. (reuters.com) Anthropic is already tied to two giants on the chip side. The company said this week it is expanding work with Google and Broadcom on “multiple generations” of custom computing systems, which suggests it wants both more supply now and more control over its hardware later. (anthropic.com) That is the real meaning of “build its own chips.” Anthropic would not be opening a factory in San Francisco; it would be designing chips the way a company designs its own engine, then relying on manufacturing partners and cloud partners to turn that design into usable machines. (reuters.com) (anthropic.com) The pressure is not only commercial. The Guardian reported on April 10 that United States officials, including Treasury Secretary Scott Bessent, summoned major bank chiefs to discuss cyber risks linked to Anthropic’s latest model, with Federal Reserve Chair Jerome Powell also reported to be involved. (theguardian.com) So Anthropic now has three problems arriving at once: demand is exploding, chip supply is scarce, and regulators are paying closer attention to what stronger models could do. Designing its own chips would not solve the cyber-risk question, but it could give the company more control over the one bottleneck every artificial intelligence lab is now fighting over: compute. (reuters.com) (anthropic.com)
