Back to Flash News

UK AI chip company Fractile completes a $720 million Series C round, valuing it at $7 billion

According to Reuters and other reports, UK AI chip startup Fractile has completed a $720 million Series C financing round, co-led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, with a post-money valuation of about $7 billion. Other investors include CapitalG, BlackRock, DST Global, the UK sovereign AI fund and the British Business Bank.

The company was founded by Walter Goodwin and is headquartered in London and Bristol, focusing on AI inference acceleration chips. Its core idea is to directly stack DRAM on top of logic chips to increase memory bandwidth. The company said this architecture can deliver about 20 times the bandwidth of HBM and more than 10 times the capacity of SRAM, addressing the trade-off between large capacity and high speed in current chips. The financing will support the shipment of the first chips and systems next year.

The UK government said the investment would help increase the UK’s share in AI hardware. Fractile takes a full-stack approach, including a machine learning compiler. The company has not yet shipped production chips.

Original source: reuters.com/legal/transactional/uk-backed-ai-chip-startup-fractile-hits-7-billion-valuation-funding-round-2026-11-11

ABAB AI Insight

AI chip startups are using memory bandwidth rather than pure compute as their core differentiator, and are securing large-scale financing and government support. Fractile completed a $720 million Series C round at a $7 billion valuation, led by several top institutions, with participation from the UK sovereign fund. The company stacks DRAM on logic chips, aiming to break the bandwidth-capacity trade-off between HBM and SRAM and serve long-context and agentic inference.

Unlike mainstream HBM solutions, this path combines high bandwidth and high capacity on the same chip. In a similar situation, capital is more focused on the mass-production feasibility of stacked packaging and next year’s shipment timeline, as well as the room for competition with NVIDIA’s dominant market position. The company emphasizes full-stack control and not hard-coding transformers into silicon to cope with rapid model iteration.

This financing will move Fractile from early-stage R&D to the preparation stage for chip shipments. The beneficiaries are cloud providers and model companies seeking alternative inference hardware; the existing HBM supply chain faces potential substitution pressure. The structural shift is that AI acceleration is moving from a compute-scaling path to a memory-bandwidth-dominant path, while the UK is increasing its domestic hardware share through its sovereign fund.

Cognitive laws

  1. Stacking DRAM directly on logic chips turns the bandwidth-capacity trade-off into a single-chip solution.

  2. Large-scale financing and sovereign-fund participation support non-mainstream architectures in moving from R&D to shipment preparation.

  3. Not hard-coding a specific model structure into silicon allows chip lifecycles to match the speed of model iteration.