Anthropic is putting together a team to design its own custom chips for AI, a move that pulls the Claude maker deeper into the hardware race. Business Insider broke the story first, and Anthropic has since confirmed the plans to TechCrunch AI. According to TechCrunch AI, the company says it wants to co-design hardware and models together so its technology runs faster and more efficiently.
This is a big signal about where the AI industry is heading. Building silicon is one of the hardest, most expensive things a tech company can attempt. When a model lab decides to do it anyway, it’s telling you that renting compute from someone else no longer cuts it.
What Anthropic is doing
The company is hiring engineers for what a job listing calls its “custom silicon team,” per TechCrunch AI. The goal isn’t just a chip for the sake of a chip. Anthropic wants to design the hardware and the models side by side, so each is tuned to the other.
A few key details from the reporting:
- The partner angle: Last month, The Information reported that Anthropic was scouting Samsung as a possible manufacturing partner, TechCrunch AI notes.
- The existing deals: Anthropic has already signed agreements with AWS, Google, Nvidia, and AMD to access AI computing hardware.
- The reason: Demand for Claude keeps climbing, and every AI company is scrambling to lock up as much infrastructure as it can.
Why this matters
Right now, most of the AI world runs on Nvidia. That dependence has created shortages, bidding wars, and eye-watering prices for the chips that train and run these models. Owning your own silicon is a way out of that trap. It means lower costs at scale, supply you control, and hardware shaped around your exact workloads instead of someone else’s general-purpose design.
What stands out here is the timing. Anthropic already has hardware deals with four major providers. If those weren’t enough, that tells you just how steep the demand curve for Claude has gotten. As TechCrunch AI puts it, relying on others clearly isn’t enough to scale to the level Anthropic needs.
Anthropic isn’t alone
This is a pattern now, not a one-off. Anthropic joins a growing list of AI heavyweights that decided to build their own chips rather than buy everything off the shelf. Per TechCrunch AI:
- OpenAI unveiled its Broadcom-built Jalapeño chip in June, designed specifically for inference workloads.
- Google DeepMind has long leaned on Alphabet’s TPU chips to power its models.
- Meta has been developing its own MTIA accelerators for AI workloads.
Anthropic is the latest to reach the same conclusion. The big labs have realized that whoever controls the hardware controls their own cost structure, their own speed, and their own destiny. Custom silicon is becoming table stakes for anyone operating at frontier scale.
There’s a useful distinction worth flagging. OpenAI’s first chip targets inference, the part where a trained model actually answers your prompts. That’s the workload that scales with users, and it’s where custom silicon pays off fastest. Anthropic’s framing, co-designing hardware and models, suggests it’s thinking about the full stack, not just one slice.
What comes next
Don’t expect an Anthropic chip tomorrow. The company is still hiring the team, and designing custom silicon takes years from job listing to working product. What you should watch for is the shape of the effort: whether Anthropic locks in Samsung or another foundry, whether it targets inference first like OpenAI did, and how much it leans on its existing partners in the meantime.
For practitioners and businesses building on Claude, the near-term takeaway is stability. A lab investing in its own hardware is a lab planning to serve a lot more demand, more cheaply, for a long time. That’s good news if your product depends on Claude staying fast and available.
The race to own the full AI stack just got one more serious contender. More details are available at the original source.