Artificial intelligence research company Anthropic is assembling an in-house semiconductor engineering team to design custom chips for its AI models, according to recent job listings and reports confirmed by TechCrunch.
The move marks Anthropic’s first formal step into proprietary hardware design, aligning its infrastructure strategy with industry peers like OpenAI, Meta, Google, and Microsoft, all of which are actively pursuing custom silicon to offset rising compute costs and hardware availability constraints.
Building Custom Hardware for Claude
Anthropic’s recent job openings focus on recruiting senior hardware engineers across key disciplines, including front-end design, pre-silicon verification, physical design, and packaging. The roles explicitly call for technical leads with experience taking semiconductor designs from initial architecture through production shipping.
By designing its own accelerators, Anthropic aims to tailor hardware specifications, such as memory bandwidth and cache hierarchies, to the requirements of its Claude model family. This co-design approach allows AI developers to optimize chip architecture for specific neural network workloads, improving processing efficiency and lowering operational costs for large-scale model inference and training.
A Multi-Vendor Compute Strategy
While Anthropic is initiating its own hardware design efforts, the company confirmed that custom silicon will complement rather than replace its existing infrastructure setup.
Anthropic maintains major cloud and hardware partnerships across the technology sector:
- Amazon Web Services (AWS): Utilizing AWS Trainium and Inferentia chips.
- Google Cloud: Leveraging Google’s Tensor Processing Units (TPUs).
- Nvidia & AMD: Utilizing standard GPU clusters for core research and operations.
Building internal chip design capabilities provides Anthropic with additional flexibility while reducing sole dependence on third-party hardware supply chains.
Part of a Broader Industry Trend
The decision reflects a broader shift across the AI sector, where leading research labs are seeking greater control over their underlying compute infrastructure.
OpenAI recently partnered with Broadcom to develop custom chips expected in 2026, while tech giants Google and Meta continue to iterate on their respective TPU and MTIA architecture programs. As model parameters grow and real-world deployment scales, custom silicon development has increasingly become a standard component of long-term AI infrastructure planning.
