Anthropic Assembles Custom Chip Team, Hires Ex-OpenAI and Google TPU Experts

Anthropic is accelerating its push into custom silicon by assembling an in-house chip design team, marking a strategic shift toward greater control over its foundational computing infrastructure. The company recently confirmed the initiative, offering compensation packages up to 485,000 USD annually for senior hardware engineers, alongside specialized roles for technical program managers and system architects. The recruitment requirements emphasize hands-on experience in frontend design, physical layout, verification, and successful tape-out processes, underscoring Anthropic’s intent to manage the entire silicon development lifecycle internally. The team expansion is being led by Amir Salek, a former Google chip executive who recently joined Anthropic’s compute organization under James Bradbury. Salek brings extensive experience from his tenure at Google, where he was instrumental in developing the first seven generations of Tensor Processing Units, as well as prior senior engineering roles at Nvidia. His arrival follows the June departure of Clive Chan, an early architect of OpenAI’s custom chip program, whose expertise in accelerator software and data center co-design complements Anthropic’s hardware ambitions. Rather than replacing external infrastructure, Anthropic’s chip strategy operates in parallel with substantial third-party commitments. The company has secured multi-gigawatt computing supply deals with Google, Broadcom, and Amazon Web Services, including a promise to invest over 100 billion USD in AWS technology over the next decade. The self-developed silicon is intended to eventually optimize Anthropic’s core infrastructure and reduce long-term dependency, while immediately augmenting its expanding external compute footprint. A distinctive feature of Anthropic’s hardware initiative is the direct integration of its large language model into the design workflow. The company is recruiting a research engineer focused on reinforcement learning for chip design, with compensation ranging from 500,000 to 850,000 USD annually. This role will establish environments for training Claude to automate tasks such as register-transfer language generation, formal verification, and physical design optimization. Concurrently, existing Claude models are already being deployed to assist current engineers in reviewing RTL code, analyzing verification coverage, and ensuring specification compliance. This dual approach of building internal silicon expertise while leveraging AI-driven automation aligns with broader industry trends toward specialized accelerators and democratized hardware development. By combining veteran chip architects with generative AI workflows, Anthropic aims to streamline hardware iteration cycles and maintain competitive advantage in the rapidly scaling AI infrastructure landscape.