Anthropic is reportedly in early discussions with UK-based semiconductor startup Fractile to secure a future supply of specialized AI inference chips. These chips are designed to run trained AI models more efficiently, helping reduce cost and improve speed. The discussions are still at an early stage, and no agreement or timeline has been finalized yet, reports The Information. Fractile’s chips are expected to launch later in the decade, suggesting this could be a long-term deal.
Inference chips like those being developed by Fractile are specially designed to run AI models efficiently after they have been trained. Unlike general-purpose GPUs, they are designed to handle repetitive, high-throughput workloads with lower energy consumption and reduced latency. This makes them particularly attractive for applications where responsiveness and cost per query are critical, like enterprise AI tools, customer service automation, and large-scale API deployments. As AI adoption accelerates across industries, demand for such specialized hardware is expected to grow significantly.
For Anthropic, the move can also be seen as part of a strategy to reduce dependence on dominant chip suppliers like Nvidia. Nvidia’s GPUs currently power much of the global AI infrastructure, but their high cost and limited supply have become constraints for many companies. By exploring alternative hardware providers, Anthropic may be aiming to diversify its supply chain, gain pricing leverage, and ensure more predictable access to compute resources as it scales its Claude family of AI models.
The timing of these discussions aligns with a wider industry trend toward custom AI silicon. Major technology companies, including Google, Amazon, and Microsoft, have already developed in-house chips customized for AI workloads, like inference accelerators and training processors. These efforts are driven by the need to optimize performance, reduce long-term costs, and gain tighter control over the hardware-software stack. And for independent AI labs like Anthropic, partnerships with specialized chip startups may offer a similar path without the need for full in-house development.
However, the report also suggests that Fractile’s chips may not reach commercial readiness until later in the decade, potentially around 2027, indicating that any arrangement would be a long-term strategic bet rather than an immediate procurement move.