Anthropic in Talks to Buy AI Chips From UK Startup Fractile
Anthropic is in early discussions with London-based semiconductor startup Fractile to secure a future supply of specialized AI inference chips, in a move that signals the Claude maker’s growing urgency to reduce its reliance on dominant chip suppliers. The Anthropic Fractile AI inference chips talks were first reported by The Information, citing people familiar with the matter.
Anthropic is considering adding Fractile as a new supplier alongside existing providers Google, Amazon, and Nvidia. The company has already taken steps to diversify its hardware stack, recently signing a deal to purchase a large volume of Google’s TPUs on terms that allow deployment outside Google Cloud data centers. The Anthropic Fractile AI inference chips arrangement, if finalized, would extend that strategy further into specialized silicon.
Why Inference Chips Are Becoming Critical for Anthropic
Anthropic’s gross profit margin for AI product operations fell short of its target last year due to higher-than-expected inference costs, according to The Information. That pressure is driving the company toward chips built specifically for running not training AI models. Fractile’s chips rely on static random access memory, or SRAM, a different memory architecture from Nvidia’s GPUs, which allows them to process AI workloads faster and with lower energy consumption.
Fractile’s chips expected to reach commercial readiness around 2027, meaning any deal would represent a long-term strategic commitment rather than an immediate procurement move. The talks remain at an early stage and a deal may not be reached. Still, the discussions reflect a broader industry shift, with AI companies actively seeking alternatives to Nvidia as annual chip and server spending climbs into the tens of billions of dollars.
