Anthropic Builds an In-House Chip Design Team to Power Claude With Custom Silicon

Anthropic has confirmed plans to develop its own custom artificial intelligence chip, becoming the latest AI lab to conclude that relying on outside hardware suppliers is no longer enough to keep pace with runaway demand.

The company disclosed that it is assembling an in-house "custom silicon team" to co-design a processor alongside its future large language models, according to a job listing and confirmation reported by TechCrunch, Reuters, Quartz, and Business Insider. The move signals that Anthropic intends to take greater control over the hardware stack that powers Claude, its flagship model family. For those tracking the AI industry coverage, the decision fits a clear pattern: every major frontier lab is now racing to own a piece of the silicon that runs its models.

Co-Designing Chips for Claude

The details point to a co-design strategy rather than a quick off-the-shelf purchase. As SiliconANGLE reported, a spokesperson told Business Insider that Anthropic will co-design the processor with its future large language models. Co-design initiatives typically focus on tailoring a chip to a specific workload — the particular calculations and memory patterns a model relies on — and such customization can significantly increase hardware efficiency.

Off-the-shelf AI accelerators are built to serve a wide range of customers and workloads, which means they carry compromises. A chip tuned specifically for Claude's architecture could, in theory, deliver more performance per dollar and per watt than general-purpose hardware. That matters enormously at the scale Anthropic operates, where even small efficiency gains translate into massive cost savings.

Hiring Engineers, Not Just Buying Chips

TechCrunch reported that Anthropic is seeking engineers with experience in chip design for its custom silicon team, per the job listing. The article noted that the company has been updated to include confirmation from Anthropic, making the plan official rather than speculative.

The hiring push comes as demand for Claude rises and AI companies scramble to lock in as many infrastructure deals as they can. For its part, Anthropic has inked agreements with AWS, Google, Nvidia, and AMD to access AI computing hardware. But to truly scale and meet the level of demand, relying on others is clearly no longer sufficient — a calculus that has now pushed Anthropic toward designing its own silicon.

This is a notable strategic shift. Anthropic has long positioned itself as a research- and safety-focused lab, but the chip announcement shows it is also building out the kind of heavy infrastructure usually associated with the largest hyperscalers.

Joining a Crowded Field

Anthropic is not the first AI company to decide that building its own chip is worth the enormous cost and complexity. The company is following a well-trodden path blazed by its largest rivals.

In June, OpenAI unveiled its Broadcom-built "Jalapeño" chip, which is designed specifically for inference workloads — the stage where a trained model actually answers user queries. Google DeepMind has long relied on Alphabet's custom Tensor Processing Units (TPUs) to power its Gemini models, giving it one of the most mature in-house silicon programs in the industry. Meta, meanwhile, has been developing its own MTIA accelerators for AI workloads.

The trend extends beyond the model labs. Cloud providers like Amazon, with its Trainium and Inferentia chips, have also invested heavily in custom silicon to reduce their dependence on Nvidia. The common thread is a desire to escape the constraints of a single dominant supplier and to capture more of the value chain.

Why Custom Silicon Now?

The economics driving this wave are straightforward. Nvidia's GPUs command enormous margins because demand for them vastly exceeds supply, and frontier AI models require staggering amounts of compute to train and serve. Every lab that depends on those GPUs is effectively paying a tax to a single vendor while competing for the same finite supply.

Designing a custom chip is an expensive, multi-year bet — it can cost hundreds of millions of dollars and require rare engineering talent — but it offers three advantages that labs increasingly see as existential. First, it can lower long-run inference costs, which are the dominant expense for a popular model like Claude. Second, it reduces strategic vulnerability to supply shortages and export controls. Third, it creates a defensible moat: a model and a chip co-designed for each other are harder for competitors to replicate.

Anthropic's decision also reflects its deepening relationship with cloud and compute partners. The company has already secured large-scale financing for AI infrastructure, and an in-house chip program is the natural next step in turning that financing into dedicated, optimized capacity.

What Comes Next

Building a competitive AI chip typically takes years from design to volume production, so Anthropic's custom silicon is unlikely to power Claude in the immediate term. The company will continue to depend on Nvidia, AMD, and cloud-provider accelerators while its new team works toward a first tape-out.

Still, the announcement is a meaningful signal of intent. It places Anthropic firmly in the same hardware race as OpenAI and Google, and it suggests that the next phase of competition among frontier labs will be fought not just over algorithms and data, but over the silicon beneath them.

Stay Ahead of AI

The battle for AI hardware is reshaping the entire industry. Keep up with every major chip launch, funding round, and infrastructure deal at our latest AI developments hub.

Read more AI news →