Anthropic Eyes UK Startup Fractile To Solve AI Memory Shortage

Anthropic, the developer behind the Claude AI model, is reportedly in early discussions with British startup Fractile to source specialized hardware for AI inference. The move signals a strategic shift as AI companies look for ways to bypass the high costs and physical limitations of traditional high-bandwidth memory. Fractile’s approach relies on Static Random Access Memory (SRAM) integrated directly into the processor, eliminating the need for separate DRAM chips that often create performance bottlenecks and supply chain headaches.

By diversifying its hardware portfolio, Anthropic aims to shield its operations from the "pricing crunch" and persistent shortages associated with industry giant Nvidia. While Anthropic already utilizes hardware from Nvidia, Google, and Amazon, adding a startup like Fractile to its supply chain suggests a growing appetite for custom, high-efficiency silicon that is purpose-built for the intensive demands of large language model inference.

The potential partnership highlights a growing trend of "de-bottlenecking" AI infrastructure. As the industry moves toward more complex models, the ability to process data without relying on expensive, scarce memory components could become a significant competitive advantage. For Fractile, securing a major customer like Anthropic would represent a massive validation of its "DRAM-less" architecture in an increasingly crowded market.

Industry analysts will be watching closely to see if these early talks transition into a formal procurement agreement and how Fractile’s chips perform in real-world environments compared to established cloud hardware. If successful, this partnership could set a precedent for other AI labs to look beyond traditional semiconductor giants for their infrastructure needs. This report was first published by Tom's Hardware.