Anthropic has begun exploring the possibility of building its own AI chips. The company has not yet made a final decision, but is already looking for a way to be less dependent on third-party infrastructure, which is needed to train the models and run Claude.
Anthropic has begun exploring the possibility of building its own AI chips. The company has not yet made a final decision, but is already looking for a way to be less dependent on third-party infrastructure, which is needed to train the models and run Claude.
According to Reuters, the project is in its early stages: Anthropic has not yet approved a specific design for the future chip and has not even assembled a separate team for this work. The agency’s sources say that the company may still abandon this idea and remain a buyer of ready-made solutions, rather than a manufacturer of its own hardware.
The reason for the interest in its own chips is simple: Anthropic is growing rapidly, and with it the need for expensive computing. According to Reuters, the company’s annual revenue has already exceeded $30 billion, while at the end of 2025 this figure was at about $9 billion. For the developer, this means not only more customers, but also a sharp increase in the need for chips for training models, processing requests and supporting infrastructure.
Anthropic is not currently relying on a single vendor. The company uses, among other things, Google’s tensor processors, as well as Amazon chips to develop and run Claude. In addition, this week Anthropic signed a long-term agreement with Google and Broadcom to further develop these processors. Reuters adds that this agreement is connected to an overall plan to invest $ 50 billion in strengthening computing infrastructure in the United States.
Developing your own chip doesn’t seem like a cheap venture. According to Reuters sources, creating a modern AI chip can cost about $500 million. Such costs are associated not only with design, but also with hiring expensive engineers and bringing production to a level where chips can be produced without a high percentage of defects. That’s why Anthropic is not rushing to make a final decision yet and is only weighing whether the game will justify such an investment.
Anthropic is not the only one looking in this direction. Reuters writes that Meta and OpenAI are also working on their own AI chips. For large model developers, it is no longer a question of prestige, but of control over a key resource: those who have their own chips or at least long-term access to them have a better chance of not slowing down in the AI race.
Previously, dev.ua wrote about how, according to Reuters, the Chinese company ByteDance, which owns the TikTok service, is developing an artificial intelligence chip and is in talks with Samsung Electronics about its production.
