Accel spearheaded the funding round for the London chip startup, with Pat Gelsinger joining as an angel investor, shortly after reports of Anthropic exploring its potential as a customer.
Fractile, a London-based startup developing inference chips that combine compute and memory on a single die, has raised $220 million to advance its hardware to production, the company announced on Tuesday.
The round surpasses the $200 million target the company was reportedly considering in late March, as noted by Electronics Weekly, elevating Fractile to a position among European chip companies presenting themselves as alternatives to Nvidia for inference processing.
The investor profile is noteworthy for this round. Accel is believed to have led, with former Intel CEO Pat Gelsinger participating as an angel investor and operating adviser.
The đź’ś of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
Existing backers Kindred Capital, the NATO Innovation Fund, and Oxford Science Enterprises, which co-led Fractile’s $15 million seed round in July 2024, also participated.Â
The technological argument challenges the prevailing architecture. Standard AI accelerators, such as Nvidia’s H- and B-series GPUs, separate the compute die from high-bandwidth memory, incurring an energy and latency cost from transferring data between them.
Fractile’s approach performs the matrix multiplications that dominate transformer inference inside SRAM cells adjacent to the compute logic, an in-memory-compute strategy that the company claims eliminates the majority of the DRAM dependency, currently a major constraint on inference costs.
Fractile boasts that its resulting chip can operate frontier models up to 100 times faster and 10 times cheaper than current GPU setups. Recent investor materials describe it as 25 times faster at one-tenth the cost.
The central technical challenge is whether these numbers hold up under production loads. So far, the company has shared simulation and small-silicon results rather than at-scale benchmarks against existing GPU clusters.
Fractile’s first commercial chip is anticipated to be available by 2027, according to the company’s public statements. The $220 million is intended to support the design through tape-out, software-stack development, and early customer integration, rather than full production ramp.
The timing of the round aligns with potential customer opportunities. Anthropic is in preliminary discussions to purchase Fractile chips once available, according to reports earlier this month.
If the relationship solidifies, Fractile would become Anthropic’s fourth named compute supplier alongside Nvidia, Google’s TPUs, and Amazon’s Trainium and Inferentia components.
Anthropic has also been considering developing its own custom AI chips, but the pursuit of Fractile suggests a continued interest in a multi-supplier strategy.
Fractile is part of a niche group of European chip startups proposing that the inference market is fundamentally different from training, and therefore conquerable. </