Anthropic Hires Former Google TPU Lead Amir Salek to Develop Custom AI Chips
Anthropic has hired Amir Salek, a former founding executive of Google's TPU project who led the development of its first seven generations of chips, to join its compute team. This move signals Anthropic's transition toward designing its own custom AI hardware. As AI models scale, top-tier labs are increasingly designing custom silicon to reduce their reliance on Nvidia and mitigate supply chain constraints. Custom hardware also allows these companies to optimize performance and efficiency specifically for their proprietary algorithms. Anthropic currently sources its chips from Nvidia, Google, and Amazon, but has recently begun recruiting for roles to build out its in-house chip division. This aligns with similar industry moves, such as OpenAI partnering with Broadcom to develop its custom "Jalapeño" inference chip.
## BACKGROUND
Tensor Processing Units (TPUs) are application-specific integrated circuits (ASICs) developed by Google to accelerate machine learning workloads, specifically matrix multiplications. Major AI developers are increasingly pursuing custom ASICs to bypass the high costs and limited availability of general-purpose GPUs.