Anthropic is considering creating its own artificial intelligence processors, according to three people briefed on the company's deliberations, as the startup and other major AI developers grapple with shortages of the specialized chips needed to train and run more advanced models.
Those familiar with the discussions say the plans remain nascent. Two people with direct knowledge and a third person familiar with Anthropic’s thinking told reporters that the company has not committed to a specific chip design and has not yet assembled a dedicated team to pursue an in-house processor. The sources said Anthropic could ultimately decide to continue purchasing chips rather than designing its own.
A spokesperson for the San Francisco-based lab declined to comment.
Anthropic’s AI assistant Claude has seen rapidly rising demand in 2026. The company said earlier this week that its run-rate revenue has climbed to more than $30 billion, up from roughly $9 billion at the end of 2025. To develop and operate Claude, Anthropic currently uses a mix of processors, including tensor processing units - TPUs - designed by Alphabet’s Google, as well as chips supplied by Amazon.
Earlier in the week Anthropic disclosed a new long-term agreement with Google and Broadcom, which contributes to the design work behind Google’s TPUs. That arrangement is tied to Anthropic’s stated intention to invest $50 billion to bolster U.S. computing infrastructure.
The conversations at Anthropic echo a broader trend among large technology firms exploring bespoke silicon for AI workloads. Designing an advanced AI chip is an expensive undertaking; industry sources estimate development costs can run roughly $500 million, reflecting the need for specialized engineers and extensive validation to prevent manufacturing defects.
Those involved in Anthropic’s internal discussions emphasized the exploratory character of the effort. No final architecture has been selected and no team has been formally chartered, the sources said. The company continues to rely on externally produced chips while assessing whether in-house design would yield sufficient advantages to justify the time and investment required.
The situation highlights the tension between rapid demand growth for AI compute and the capital- and skill-intensive nature of chip design. Anthropic’s recent commercial deals and its pledge of a multibillion-dollar computing investment illustrate one path companies are taking to secure capacity, while consideration of an in-house chip underscores the alternative of vertical integration if supply constraints persist.
Summary
Anthropic is weighing the option of designing its own AI chips as part of a preliminary review, but it has not yet committed to any design or assembled a team. The company currently uses Google TPUs and Amazon chips, has signed long-term agreements with Google and Broadcom, and reported a significant increase in run-rate revenue for its Claude model.