UK chip startup Fractile seeks $200M round at $1B valuation to take on Nvidia in AI inference

What's the deal? Fractile, a London-based AI chip startup, is in talks to raise more than $200 million at a $1 billion valuation. The round is being discussed with Accel, the NATO Innovation FundDealroom has a profile for this one. Try Dealroom →, and Oxford Science EnterprisesDealroom has a profile for this one. Try Dealroom →, though no deal has been finalised and terms could still change.

Founded in 2022 by Walter GoodwinDealroom has a profile for this one. Try Dealroom →, a former Oxford PhD, Fractile builds chips designed specifically for AI inference — the process of running trained models — using static random-access memory (SRAM) rather than the traditional GPU architecture dominated by Nvidia.

Why now? Investor appetite for Nvidia alternatives has been rekindled by growing evidence that GPUs face structural limitations in inference workloads. Nvidia itself acknowledged the shift this month by launching a dedicated inference chip, following its $20 billion deal with Groq in December. The inference market is expanding rapidly as companies move from training models to deploying them at scale.

The UK government's push for domestic AI capabilities adds another tailwind. Last month Fractile announced £100 million in planned investment over three years to expand in London and Bristol, including a new industrial hardware engineering facility — a signal the company is moving from research-stage startup toward production-scale manufacturer.

What could go wrong? The UK chip sector has a cautionary tale close to hand. Graphcore, once one of Britain's most celebrated AI chip startups, was acquired by SoftBankDealroom has a profile for this one. Try Dealroom → in 2024 for just above $600 million — less than its total venture capital raised. Building competitive chip hardware requires sustained capital, specialised talent, and customer wins against entrenched incumbents.

Fractile also faces well-funded rivals on multiple fronts: Groq, Tenstorrent, and Olix — the latter raising $220 million last month — are all targeting the same inference efficiency niche, while Nvidia and AMDDealroom has a profile for this one. Try Dealroom → are actively adapting their own stacks to defend the category.

The signal: Fractile's fundraise is part of a broader wave of inference-focused chip startups attracting serious capital as the economics of AI deployment come under pressure. Training is largely a solved problem for incumbents with scale; inference is where costs, speed, and efficiency matter most for commercial viability. Startups that can demonstrate a genuine architectural advantage in that workload are becoming attractive bets — and governments keen on sovereign AI infrastructure are increasingly willing to co-invest alongside private capital to build domestic alternatives to the US-dominated GPU supply chain.

Sources:
Financial Times
TFN
Data Center Dynamics

Image Credits:
Fractile

A.M.