Anthropic Builds In-House Team for Custom AI Chips

Anthropic Joins the Custom Silicon Race

Anthropic has confirmed it is assembling an in-house chip design team dedicated to building custom hardware for its Claude family of AI models, formalizing what had previously circulated as an unconfirmed report 1237. The company is now actively hiring engineers to staff the effort, according to job listings and a company statement described in coverage of the move 3. Reuters, cited by one outlet, reports that Anthropic is responding directly to a shortage of chips needed to both power its existing models and develop more advanced successors 7.

Why Anthropic Wants Its Own Chips

The rationale offered across the coverage is consistent: Anthropic wants to boost inference speed and improve scale efficiency for Claude as demand for its models grows 12. Rather than relying solely on third-party chipmakers, an in-house design team would let Anthropic tailor silicon specifically to the computational patterns of its own models, a strategy that mirrors moves already made by larger, better-resourced rivals. Because inference — the process of actually running a trained model to generate responses — increasingly dominates the ongoing cost of operating large AI systems, custom silicon tuned for that workload can meaningfully cut expenses and latency compared with general-purpose GPUs.

The Broader Custom-Silicon Trend

Anthropic's move fits into a wider industry pattern in which major AI players are moving away from near-total dependence on Nvidia's GPUs. Google, the most advanced example, has built its Tensor Processing Units (TPUs) for years, and one analyst projection cited in coverage suggests Google could manufacture more AI accelerators in 2028 than Nvidia sells that year, a claim attributed to Fubon Research 4. If accurate, that would mark a striking shift in the balance of power across the AI hardware supply chain, underscoring how seriously hyperscalers and well-funded AI labs are now treating in-house or custom-designed chips as a competitive necessity rather than a luxury.

That shift is also rippling through the supply chain supporting chip production. Veeco, a semiconductor equipment maker, told investors it expects $780 million to $810 million in 2026 revenue and plans to more than double its 2027 advanced packaging and silicon photonics capacity, citing AI-driven demand and roughly $200 million in advanced packaging orders 6. That kind of capacity expansion signals confidence that demand for custom and advanced AI chips — not just commodity GPUs — will keep climbing.

Labor Undercurrents

Separately, broader commentary on AI's impact on the tech workforce points to a less-discussed side effect: rising interest in unionization among engineers and technical staff, driven partly by layoffs tied to AI adoption 58. While not directly tied to Anthropic's chip plans, this trend reflects growing anxiety across the same industry now racing to build ever more specialized AI infrastructure, suggesting that the economic upheaval driving hardware innovation is also reshaping labor dynamics within the companies building it.

What It Means Going Forward

Anthropic's custom chip ambitions remain in early stages, with hiring just beginning, but the announcement signals that even AI labs without Google's or Amazon's hardware history now see proprietary silicon as essential to controlling costs and securing supply as chip shortages persist.