Anthropic is in early talks to buy artificial intelligence inference chips from Fractile, a London-based chip design startup founded by a former Oxford University PhD student, in a potential deal that would bolster the UK's position in the global AI hardware supply chain.
The Telegraph reported the discussions on Saturday, framing them as part of Anthropic's urgent search for ways to cut the cost of running its Claude models, which are currently served at a loss despite annualised revenue surpassing $30 billion.
Fractile, founded in 2022 by Walter Goodwin, designs processors specifically optimised for inference, the computationally intensive process by which AI models generate responses to user queries.
Training a model is a one-off cost, but inference runs continuously every time a user interacts with it, and as Claude's popularity has surged this year, the inference bill has become one of Anthropic's largest and fastest-growing expenses.
A deal would give Anthropic a fourth chip supply track alongside Nvidia GPUs, Google TPUs and Amazon's custom Trainium processors, diversifying a compute mix that the company has been visibly broadening as demand outstrips its existing capacity.
Fractile's chips are not expected to reach commercial readiness until approximately 2027, placing any deployment in the same window as the expanded Google-Broadcom TPU ramp that will bring multiple gigawatts of new capacity online for Anthropic.
The late-decade timeline fits a pattern the company has established: open multiple parallel chip tracks, accept that some are long-horizon hedges, and let demand decide which graduate to volume production.
In February, Fractile announced plans to invest £100 million to extend its UK operations over the next three years, in what ministers described as "a significant boost to the UK's AI hardware ecosystem."
Science and Technology Secretary Liz Kendall said last week that the government was working on an "AI hardware plan" to be unveiled in the coming weeks, and named Fractile as one of the country's brightest prospects in the field.
The talks come alongside a broader deepening of Anthropic's UK presence.
The company has secured 158,000 square feet of office space in London's Knowledge Quarter at Regent's Place, with capacity for up to 800 employees, a fourfold increase from its current 200-person team, which includes roughly 60 AI safety researchers.
Anthropic is offering London-based engineers salaries of up to £630,000 per year, a level that investors warned would force European founders to reassess their own compensation packages.
AI-linked companies have leased more than one million square feet of London office space since the start of 2025, equivalent to roughly 7% of all lettings in the period, according to Knight Frank.
OpenAI has also opened its first permanent London office in the Knowledge Quarter, while Jeff Bezos' secretive AI venture Project Prometheus is reportedly in talks to secure space nearby in King's Cross.
Related reading
- OpenAI deploys GPT-5.5 across Nvidia on Blackwell infrastructure in enterprise-scale AI agent rollout
- Robinhood Ventures invests $75m in OpenAI
- Anthropic finalising $1.5bn joint venture with Blackstone, Goldman Sachs and Hellman & Friedman
For Fractile, a supply agreement with the world's second most valuable private AI company would represent a transformative commercial validation.
For Anthropic, it would represent another hedge in an infrastructure strategy built on the assumption that no single chip supplier can deliver the scale its growth demands.
The recap
- Anthropic in talks to buy microchips from a British start-up
- Major artificial intelligence labs, including Anthropic, run systems at a loss
- Report says deal would underscore London's status as AI hub
