Anthropic in Talks to Purchase Inference Chips from UK Startup Fractile

Introduction

According to sources cited by tech outlet The Information, AI giant Anthropic is in early-stage procurement talks with UK chip startup Fractile, planning to integrate the latter's chips into its supply chain once they hit the market in 2027. The move signals Anthropic's active efforts to diversify its chip supply sources in response to the massive computing demands driven by explosive business growth.

Surging Demand Forces Supply Chain Diversification

Anthropic has risen rapidly in recent years on the strength of its Claude series of large language models, with commercial revenue growing at a breakneck pace. However, the rapidly expanding user base and surging inference request volumes are placing enormous pressure on its existing server infrastructure. Currently, Anthropic relies primarily on three major suppliers for its AI server chips — Google, Amazon, and Nvidia.

Against this backdrop, Anthropic has begun turning its attention to emerging chip manufacturers. Fractile, a UK-headquartered AI chip startup specializing in AI inference chip development, expects to bring its products to market in 2027. Although both parties are still in the early stages of discussions, the signal itself is noteworthy.

Fractile: A New Player Targeting the Inference Market

While Fractile is not yet widely known, the inference chip segment it focuses on is currently the fastest-growing niche within the AI industry chain. As large models progressively move from the training phase toward large-scale deployment and commercial applications, computing power consumption during inference is rapidly surpassing that of training, becoming the primary component of operational costs for AI companies.

For Anthropic, choosing a chip company based in the UK also carries strategic significance in terms of geographic supply chain diversification. At a time when the global chip supply chain is highly concentrated in the United States, bringing in a European supplier helps mitigate potential supply chain risks.

Industry Landscape: AI Companies Race to Reduce Nvidia Dependence

Anthropic's move also reflects a broader trend across the AI industry — leading AI companies are actively seeking chip alternatives beyond Nvidia. Google has developed its own TPU chips, Amazon has launched its Trainium and Inferentia chip lines, and Microsoft is developing its proprietary AI chip Maia. The core logic behind all these initiatives is to reduce over-reliance on a single supplier while optimizing performance and controlling costs through customized hardware.

Meanwhile, a wave of startups focused on AI inference chips is also rising rapidly, with companies like Groq, Cerebras, and d-Matrix securing substantial funding rounds. If Fractile can successfully land Anthropic as a marquee customer, it would undoubtedly secure a favorable position in the fiercely competitive AI chip market.

Outlook: The AI Chip Market in 2027

Notably, Fractile's chips are not expected to be delivered until 2027, meaning Anthropic's potential purchase is more of a forward-looking strategic move. By 2027, AI inference demand is projected to grow several-fold from current levels, and the market's appetite for high-performance, low-power inference chips will only intensify.

However, the two-year window is also fraught with uncertainty. Whether Fractile can complete chip development and achieve mass production on schedule, and how Anthropic's technology roadmap and business scale will evolve by then, are all critical variables that could influence the final outcome of the deal.

Overall, this potential transaction is not only a pragmatic move by Anthropic to address its computing power bottleneck but also a microcosm of the accelerating restructuring of the global AI chip supply chain. The competitive battle in the AI inference chip market has only just begun.