Anthropic is assembling an internal team to design its own artificial intelligence chips, according to people familiar with the effort. The team is being led by a former OpenAI engineer who previously worked on that company's chip initiatives. The move signals that Anthropic, best known for its Claude family of large language models, wants more control over the hardware that powers its AI.
Why build in-house?
Right now, nearly every AI company relies on chips from Nvidia or, to a lesser extent, AMD and Intel. Those chips are expensive and often in short supply. By designing its own silicon, Anthropic could tailor the hardware to its specific models, potentially cutting costs and improving performance. The company hasn't said which chip designs it's pursuing, but the team is expected to focus on accelerators for training and inference.
The move also reduces dependence on a single supplier. Nvidia's GPUs are the gold standard for AI, but they're hard to get and come with a premium. Anthropic's in-house effort could give it more leverage in negotiations and a fallback if supply tightens further.
An OpenAI veteran at the helm
The person leading the new team spent years at OpenAI, where they worked on that company's early chip projects. OpenAI itself has explored building custom chips, though it has also partnered with Microsoft and others. Anthropic's hire suggests it's serious about making its own silicon a reality, not just a research project.
Anthropic declined to comment on the team's size or specific goals. But the company has been on a hiring spree for hardware engineers, posting job listings for chip architects and ASIC designers in recent months.
What this means for the AI landscape
If Anthropic succeeds, it could reshape the competitive dynamics of the AI industry by 2026. Custom chips could give Anthropic's models a speed or cost advantage over rivals that rely on off-the-shelf hardware. That might also make Anthropic a more attractive partner for cloud providers and enterprise customers who want optimized AI infrastructure.
But building chips from scratch is notoriously hard and expensive. Even Google, with its Tensor Processing Units, took years to get them right. Amazon's AWS has its Trainium and Inferentia chips, but they still lag behind Nvidia in many benchmarks. Anthropic will need deep pockets and patience.
The company has raised billions from investors including Google and Amazon, so it has the capital. The question is whether it can execute fast enough to matter before the next wave of Nvidia hardware arrives.
Anthropic is expected to start hiring aggressively for the chip team in the coming months. The first prototypes, if they come, are likely still a couple of years away. In the meantime, the company will continue to rely on Nvidia's GPUs for training its models.
The chip effort is one of several moves Anthropic is making to secure its future. It has also been building out its own cloud infrastructure and exploring new model architectures. Whether the chip team delivers or not, the message is clear: Anthropic wants to own more of its stack.
