Anthropic is hiring an AI chip design team
TechCrunch Rebecca Bellan
Anthropic is building an in-house team to design its own AI chips, per Business Insider. It's chasing OpenAI, Google, and Meta into the custom-silicon race.
Anthropic wants to stop being a customer and start being a chipmaker, at least partly. Business Insider reports the Claude developer is staffing up a "custom silicon team," with job listings seeking engineers who've actually built chips before, not just used them. The stated goal is co-designing hardware alongside its models so Claude runs faster and cheaper, rather than bolting software onto whatever GPU happens to be available that quarter.
This isn't a sudden pivot. The Information reported last month that Anthropic had been in talks with Samsung about manufacturing custom chips, suggesting the groundwork here goes back months. And it makes sense given where Anthropic sits right now: demand for Claude keeps climbing, and the company has spent the past year signing compute deals with AWS, Google, Nvidia and AMD just to keep pace. Diversifying suppliers only gets you so far when everyone else is fighting for the same limited pool of advanced chips.
Anthropic is late to this particular party, not first. OpenAI unveiled Jalapeño, its Broadcom-built inference chip, back in June. Google has run its own TPUs for years through DeepMind, and Meta has been quietly iterating on MTIA accelerators for its own workloads. What's notable is less that Anthropic is doing this and more that doing it has become table stakes. If you're serious about frontier AI at scale, apparently you eventually need your own silicon roadmap, not just a purchase order.
Anthropic didn't respond to a request for comment, so the details—timeline, partners, whether Samsung is actually in the picture—remain thin. But the direction is clear enough: the biggest AI labs are deciding that renting infrastructure from Nvidia and the cloud giants is a tax they'd rather stop paying, even if building your own chips is slower, riskier, and vastly more expensive up front.
My take
Every major lab building its own chips at once is a pretty clear signal that Nvidia's margins have become everyone else's problem, and this is the predictable response. It also quietly answers the open-vs-closed debate in a way people don't talk about enough: vertical integration like this favors whoever already has billions in committed cloud revenue, which is exactly the kind of moat that makes it harder for smaller or open-source-oriented labs to compete on raw compute terms.
Read more about this at: TechCrunch
