Magazine Artificial Intelligence

Fractile raises $720M Series C to challenge Nvidia in AI chips

Fractile raises $720M Series C to challenge Nvidia in AI chips

London startup Fractile develops advanced chips and systems for artificial intelligence. It has closed a Series C round of €649 million, equivalent to $720 million. The round was co-led by Thrive Capital, Founders Fund, Lightspeed and Redpoint. Among the investors is also CapitalG, Google’s venture capital arm. BlackRock, DST Global, Elad Gil, Balderton and SV Angel complete the list. The UK sovereign fund for artificial intelligence and the British Business Bank are also taking part.

According to Reuters, the round also includes substantial public backing from the British government. AI minister Kanishka Narayan described the deal as «a massive act of confidence in British technology». He also called it confidence «in our ambition to reindustrialize Britain».

What the AI inference chip is for

Fractile was founded in 2022 by Walter Goodwin and Yuhang Song, with headquarters in London and operations in Bristol. Goodwin, the company’s CEO, conducted his doctoral research in robotics and artificial intelligence at the University of Oxford. The company develops processors dedicated to AI inference, the process by which trained models generate responses and predictions. With the rise of reasoning models and autonomous agents, demand for inference compute is growing.

The startup’s core insight is precise. The next limit of advanced AI systems will not be their intelligence. Instead, it will be the time and cost needed to produce useful output. Fractile’s technology targets thousands of tokens per second, even on large models with extended context windows. If it achieves those speeds, AI systems could complete complex reasoning tasks faster and reduce the operating costs of the most sophisticated models.

Three-dimensional memory at the heart of the architecture

The technical design centers on a three-dimensional architecture that stacks DRAM memory directly on top of the processing logic. This design aims to reduce the time and energy required to move data between memory and compute components. That bottleneck currently limits the performance of existing accelerator chipsets.

According to Fractile, the architecture can deliver substantially higher memory bandwidth than conventional high-bandwidth memory (HBM), while at the same time providing greater capacity than SRAM. Phil Chambers, partner at Balderton, explained the team’s bet: «Walter and the Fractile team made a bold bet in 2022: that the real limit of frontier AI would be the speed at which it produces useful output, and that the only way to solve that at scale was to rebuild the hardware itself».

He said it controls every layer, from silicon to systems, software and fundamental ML research. «They have built an architecture that delivers speed and scale at the same time».

The British government backs the chip ecosystem

The deal fits into a series of European investments in AI compute hardware in 2026. In the United Kingdom, London startup OLIX raised €270.5 million for AI inference chips. Dutch startup EUCLYD secured over €200 million for custom silicon and compute systems. In Barcelona, Openchip attracted €115 million in public investment. Valencia-based iPronics raised €107.6 million for AI optical networking. British startups Callosum, Optalysys and Tendrils together secured €137.8 million.

Overall, these announcements represent more than €830.8 million in funding announced in 2026. The figure rises to about €1.48 billion including Fractile’s round. Minister Narayan acknowledged a historical problem for the country. «For too long Britain has produced some of the best scientific ideas in the world». He added that much of the technology, industry and wealth they create has been built elsewhere. «We are determined to change that», the minister said.

How much Fractile has raised and what remains to be proven

The new round follows a €198 million ($220 million) Series B. That round closed in May 2026. Previously, the startup had raised a €13 million ($15 million) seed round in 2024. The deal takes place in an increasingly competitive AI semiconductor market. In that market, Nvidia remains the main supplier of the processors used to train and run advanced models. Companies developing dedicated inference accelerators are trying to meet demand for faster, more energy-efficient alternatives.

For Fractile, the next milestone is the commercial launch: the first commercial chip is scheduled for 2027. The performance and efficiency gains the company claims will still have to be validated in widespread production use.

Sources: