Anthropic is in serious talks with Samsung about building a custom AI chip, according to TechCrunch. The Information first broke the story, reporting that the two companies are actively exploring a partnership around the chip project. It marks a notable step forward from April, when Reuters revealed that Anthropic was only loosely considering the idea of making its own silicon in response to ongoing chip shortages.
The details are still vague. Anthropic has not yet decided what the chip will actually do, how it will fit into a server, or how much computing power it needs to have. That said, the fact that Samsung is already at the table suggests this is more than just early brainstorming.
When asked for comment, Anthropic told TechCrunch that its compute strategy will continue to rely on a mix of chips from Google, Amazon, and Nvidia. On the Samsung talks specifically, the company said it had nothing to add.
The move fits into a broader trend across the AI industry. Several major AI companies are now working on their own custom chips for two main reasons:
- To build hardware tuned for their specific workloads, rather than relying on general-purpose chips
- To reduce their dependence on Nvidia, which still dominates the chip market by a wide margin
Anthropic’s push also comes just days after rival OpenAI announced its own custom inference chip, called “Jalapeño,” built in partnership with Broadcom. OpenAI claims the chip delivers better performance per watt than competing chips. Google and Amazon have each offered custom-built chips through their cloud platforms for some time now, so Anthropic is arguably playing catch-up.
Samsung is a logical partner for this kind of project. The company already has deep ties to Nvidia, manufacturing chips that Nvidia needs to train and run AI models. Nvidia’s software is also part of Samsung’s own chip production process. The two are even building an AI chip factory together in South Korea. Samsung has also held separate talks with Google about chip manufacturing. In short, Samsung knows this space well and has the relationships and infrastructure to back a project like this.
For Anthropic, getting custom silicon right could matter a lot down the line. As the company scales its Claude models and competes for enterprise customers, controlling more of its hardware stack gives it more options on cost, speed, and availability. It is not a quick fix, but it is the kind of long-term bet that the biggest players in AI are now treating as essential.
