Anthropic wants to build the silicon its models run on. The company confirmed it’s hiring a “custom silicon team” to design chips for Claude, which puts it on the same road OpenAI, Google and Meta are already walking.
The confirmation came the hard way, through a job board. Business Insider spotted a listing yesterday for a senior engineer with experience shipping semiconductor designs. Two listings are live right now: a silicon engineer and a technical program manager, silicon. A spokesperson for Anthropic then confirmed the plans to both Business Insider and TechCrunch.
Worth reading closely: the spokesperson said Anthropic will still take a “multi-chip approach.” That means hardware from other companies alongside its own designs as it keeps scaling. This isn’t a clean break from anyone.
The rumor was already out there
None of this came from nowhere. The Information previously reported Anthropic was considering Samsung as a hardware manufacturing partner, and the hiring push lines up with that.
OpenAI got there first with an announcement. Its custom chip, called Jalapeño, is built for large language model inference in data centers and developed with Broadcom. Google has run its models on its own hardware for a while. Meta has designed and deployed its own chips. Mistral is reportedly looking at the same move.
Nvidia is the reason, mostly
Much of the industry runs on Nvidia hardware, and that concentration is a strategic vulnerability. Compute infrastructure is fiercely contested right now, with demand still outstripping current capacity. Depending on one supplier in that market isn’t a comfortable position.
The second reason is performance. Designing chips for specific models, and models for specific chips, should produce better results than generic hardware. If OpenAI banks the returns on that vertical integration, Anthropic and every other frontier model provider will want the same thing.
Anthropic says its teams will co-design new hardware and models side by side. The company has co-designed hardware with partners before. What’s different is pulling the silicon expertise in-house.
There’s an edge-computing angle too
Developers and other users are starting to run cheaper, smaller or open-weight models on their own hardware and on edge devices. Anthropic may be hoping custom silicon gives its frontier models something those setups can’t match.
But the team doesn’t exist yet. Anthropic is still hiring the people who would do this work, and until those seats are filled, neither the company nor anyone using Claude sees a thing from it.
