Fractile Raises $220 Million to Bring Its In-Memory Compute Inference Chip into Production
May 13, 2026 – 3:18 pm
Image by: Fractile
The London-based startup designing inference chips that integrate compute and memory on a single die, Fractile, has secured $220 million in funding to advance its hardware into production. This round, led by Accel with the participation of former Intel CEO Pat Gelsinger as an angel investor, surpasses the previously targeted $200 million, as reported by Electronics Weekly.
This investment positions Fractile among European chip companies aiming to challenge Nvidia’s dominance in the inference layer of AI processing. Kindred Capital, NATO Innovation Fund, and Oxford Science Enterprises, who backed Fractile in its $15 million seed round in July 2024, also contributed to this latest funding.
Fractile’s technology challenges the conventional architecture of AI accelerators like Nvidia’s H- and B-series GPUs. Current setups separate compute dies from high-bandwidth memory, leading to energy and latency issues due to data transfer. Fractile’s innovative design conducts matrix multiplications within SRAM cells adjacent to the compute logic, enabling in-memory computing that significantly reduces reliance on DRAM—a key constraint on inference cost.
The company claims its chip can execute frontier models up to 100 times faster and 10 times cheaper than current GPU setups, or even 25 times faster at one-tenth the cost according to recent investor materials. However, the proof of these claims awaits independent benchmarks against deployed GPU clusters in production environments.
Fractile’s first commercial chip is expected to be available in 2027, and this $220 million funding will cover tape-out, software stack development, and early customer integration. Timing is particularly favorable as Anthropic, a prominent AI company, is reportedly in early discussions to become a Fractile chip customer.
If the partnership materializes, Fractile would be one of four known compute suppliers for Anthropic, alongside Nvidia, Google’s TPUs, and Amazon’s Trainium and Inferentia chips. This diversity suggests Anthropic continues to explore custom AI chip design while also considering multi-supplier strategies.