British chip upstart Fractile just landed a massive cash injection to chase one of AI’s biggest headaches: slow, costly inference. The London-based company sealed a $220 million Series B, drawing in some of the biggest names in venture capital. The deal also signals a serious push to break Nvidia’s grip on the AI hardware market. And one well-known AI lab is already circling.
Inside the $220M Series B Round
The funding round was co-led by Accel, Factorial Funds, and Peter Thiel’s Founders Fund. The round also brought in Conviction, Gigascale, O1A, Felicis, Buckley Ventures, and 8VC, alongside existing backers.
Existing investors Kindred Capital, the NATO Innovation Fund, and Oxford Science Enterprises, which co-led Fractile’s $15 million seed in July 2024, are part of the round. Accel is understood to have led, with former Intel chief executive Pat Gelsinger participating as an angel and operating adviser.
The round closes above the $200 million reported target the company was understood to be sounding out in late March, lifting Fractile into the cohort of European chip companies pitching themselves as alternatives to Nvidia at the inference layer.
| Deal Snapshot | Details |
|---|---|
| Round | Series B |
| Amount Raised | $220 million |
| Lead Investors | Accel, Factorial Funds, Founders Fund |
| Founded | 2022, London |
| Founder and CEO | Dr. Walter Goodwin |
| First Chip Delivery | Targeted for 2027 |
Why Inference Has Become AI’s Biggest Bottleneck
Inference is the moment an AI model actually answers a question. It is also where most of the running cost lives. Training will continue to require the largest, most exotic systems and Nvidia’s CUDA moat is strongest there, while inference, the workload that actually consumes most of the dollars once a model is deployed, rewards specialised architectures tuned for throughput and energy per token rather than peak FLOPs.
Founder Walter Goodwin says the math no longer works on today’s chips. “At the roughly 40 tokens per second at which these models tend to run on existing chips, a single output of this length takes a month to complete,” Goodwin said.
The company says current architectures face a memory-bandwidth bottleneck as models produce longer outputs and handle larger context windows. That single technical wall is what Fractile wants to demolish.
“Inference is both the revenue engine of the AI industry and the rate-limiting factor on expanding it.” — Walter Goodwin, Founder and CEO, Fractile
The In-Memory Compute Bet
Fractile’s pitch is built on a different chip layout than what Nvidia ships today. Founded in 2022 by Oxford PhD Walter Goodwin, Fractile is developing an inference chip that co-locates memory and compute on the same die using SRAM rather than shuttling data to separate DRAM chips.
That data movement between the GPU and off-chip DRAM is one of the main bottlenecks in running large AI models at speed. By killing that traffic, Fractile believes it can deliver huge speed and cost wins.
The performance claims are bold:
- Performance could climb from roughly 40 tokens per second on existing hardware to around 1,200 tokens per second, compressing workloads that currently take a month into a single day.
- Fractile claims the chip can run frontier models up to 100 times faster and 10 times cheaper than current GPU setups; more recent investor materials frame the comparison as 25 times faster at one-tenth the cost.
- Whether those numbers hold under production loads is the central technical question, since the company has so far disclosed simulation and small-silicon results rather than at-scale benchmarks against deployed GPU clusters.
Anthropic Is Watching Closely
One of the most striking signals for Fractile is the interest from a top-tier AI lab. Anthropic has reportedly held early discussions with London-based chip startup Fractile about purchasing the company’s inference accelerators, which would add Fractile as a fourth source of AI server silicon for the Claude developer, which already uses chips from Nvidia, Google, and Amazon.
Fractile’s chips aren’t expected to reach commercial readiness until around 2027, placing any deployment well outside Anthropic’s near-term procurement plans.
The competitive lane is getting busy. Groq has shipped its language-processing units to multiple model providers and recently raised at a $6.9 billion valuation; Etched is building transformer-specific silicon; Cerebras and SambaNova have raised against the same workload. Nvidia struck a $20 billion acquisition deal with Groq in December and subsequently launched its own dedicated inference accelerator, Groq 3 LPX, acknowledging the growing commercial pressure to optimize cost-per-token at scale.
A Win for UK Sovereign AI Ambitions
British AI chip startup Fractile has raised $220m (£165m) in fresh funding with the government pointing to the deal as evidence the UK can produce globally competitive AI infrastructure companies.
The round follows Fractile’s February announcement of a £100 million ($132 million) three-year expansion of its London and Bristol operations, including a new hardware-engineering site in Bristol, and fits the wider UK sovereign-AI push that also produced the BT, Nscale, and Nvidia data-centre partnership in April.
“AI will be critical to the UK’s future prosperity and security, and next generation AI chips like Fractile’s are a key part of making that happen,” with the government expected to publish its AI hardware plan later this year as ministers attempt to strengthen Britain’s position in the tech’s infrastructure.
The company is now in scale-up mode. The team has drawn engineers from Graphcore, Nvidia, and Imagination Technologies, and is building its software stack alongside the silicon. Fractile is now hiring across London, Bristol, San Francisco, and Taipei as it prepares for global expansion and commercial deployment in the coming years.
For an industry that runs on tokens per second and dollars per query, Fractile’s bet is simple but daring. If the in-memory design holds up in real data centres, a small British team could rewrite the economics of frontier AI. If it stumbles, the inference race will roll on without it. Either way, the next 18 months will test whether bold simulations can survive contact with production. What do you think about Fractile’s challenge to Nvidia? Share your thoughts in the comments below and pass this story along to anyone tracking the future of AI hardware.
