Anthropic Is Building Its Own AI Chips — and Hiring the Team to Do It

Anthropic is getting into the chip business.

According to Ars Technica, the company has confirmed plans to build an in-house silicon team, designing its own hardware to power Claude. TechRepublic reports that Anthropic is actively hiring engineers for the effort, and that the team is exploring custom AI chips intended to reduce compute constraints and support Claude's continued growth.

As Ars Technica frames it, Anthropic and OpenAI are both racing to scale up while reducing their dependence on Nvidia.

That last point is the heart of the story. Today, training and running large AI models means buying enormous quantities of specialized processors, and Nvidia dominates that market. For an AI company, chips are the single largest input cost and the hardest thing to get more of. When supply is tight, your ability to grow isn't set by how good your models are — it's set by how many chips a supplier is willing to sell you and at what price.

Designing your own silicon is a way out of that bind, at least partially. Custom chips can be tuned specifically for the workloads a company actually runs, rather than being general-purpose parts sold to everyone. Google has taken this path with its own AI processors; Amazon has too. What's notable here is that a model developer known for research is now planning to reach down into the hardware layer itself.

It's worth being clear about what has and hasn't been announced. The sources describe a team being assembled and custom chips being explored — not a finished product, a shipping date, or a manufacturing partner. Chip design is slow, expensive work, typically measured in years rather than quarters.

Why it matters: if AI companies start designing their own hardware, the industry's biggest bottleneck — and Nvidia's grip on it — could begin to loosen.