Amir Salek, the hardware executive who spent nearly a decade leading Google's custom silicon organization, has been reported to be joining Anthropic's compute team — a move that would place one of the tech industry's most experienced chip leaders inside the AI safety company at a moment when compute infrastructure has become the defining bottleneck in the race to build more powerful AI systems.

Salek led Google's custom silicon efforts from 2013 to 2022, a period that encompassed the development and scaling of the company's Tensor Processing Units (TPUs). Those specialized AI chips gave Google a meaningful edge in training large language models before the rest of the industry caught up. Google first unveiled the TPU in 2016, and the accelerator family has since become a cornerstone of both the company's internal AI infrastructure and its cloud offering for external customers. His tenure coincided with a broader shift among large technology companies toward designing their own accelerators rather than relying solely on outside suppliers such as Nvidia — a path since followed by Amazon with its Trainium chips, Microsoft with its Maia accelerator, and Meta with its MTIA silicon.

After leaving Google, Salek moved to the capital side of the deep tech sector, joining Cerberus Capital Management, the global alternative investment firm, as Senior Managing Director, where his work centered on deep tech investments.

For Anthropic, hardware leadership of this kind speaks to a core constraint. Founded in 2021 by former OpenAI researchers, including CEO Dario Amodei and President Daniela Amodei, Anthropic develops the Claude family of AI models and has consistently pointed to compute access as one of the key limits on frontier AI research — a challenge shared across an industry in which demand for advanced AI chips continues to outstrip supply. Anthropic, notably, trains and serves those models on silicon designed by others: Amazon has invested up to $4 billion in the company and AWS serves as its primary cloud provider, while Google, also an Anthropic investor, supplies TPUs through Google Cloud. In late 2024, Anthropic and AWS announced Project Rainier, a training cluster built on AWS Trainium2 chips — reportedly spanning hundreds of thousands of them — which the companies described as one of the largest compute clusters in the world.

The reported appointment has not been independently confirmed, however. As of August 2026, there has been no widespread corroborating reporting on Salek joining Anthropic, and the company has not publicly acknowledged any association with him in its communications.