Fractile Raises £540 Million for AI Inference Chips at $7 Billion Valuation
In less than six months, British startup Fractile has leapt from a $1 billion valuation to $7 billion. The £540 million ($720 million) Series C round, announced on Monday (November 11), places the company among the most valuable UK technology startups — alongside names like Revolut.
The investment, backed by UK government funds Sovereign AI and British Business Bank, reflects a clear bet: the bottleneck for the next generation of artificial intelligence is not raw computational power but memory bandwidth.
The thesis behind the architecture
Since its founding in 2022, Walter Goodwin and the Fractile team have operated on a premise that challenges industry consensus. While the sector continues investing in greater transistor density and higher raw compute, Goodwin argues that the true limiting factor for AI models is the speed at which data moves between memory and processors.
Fractile's solution stacks DRAM chips directly on top of logic chips — a 3D approach designed to surpass what the company considers inadequate in current high-bandwidth memory (HBM) solutions. The goal is to deliver simultaneously higher bandwidth than HBM and substantially greater capacity than static SRAM-based architectures.
According to the company, this approach would enable accelerators to run the most advanced AI models at more than ten times faster speeds than current solutions, with significantly reduced energy consumption and operating costs.
Why the valuation jumped so much
The leap from $1.1 billion in May to $7 billion in November is atypical — even by the frenetic standards of the AI market. Investors appear to have been convinced by prototypes and by the team's capability, which Goodwin assembled by recruiting veterans who have shipped silicon at scale. Fractile operates offices in London, Bristol, Palo Alto, and Taipei — unusual for a British hardware startup.
The investor lineup appears carefully constructed to reassure both the market and the government: Thrive Capital, Founders Fund, Lightspeed, and Redpoint led the round, with participation from CapitalG (Alphabet's venture arm), BlackRock, DST Global, Bain Capital Ventures, T. Rowe Price, SV Angel, Shopify Ventures, and angel investor Elad Gil. Returning investors include Accel, Conviction, Cocoa, Factorial, Felicis, Gigascale, and Kindred.
But the political signal is equally important. The participation of the Sovereign AI Fund and the British Business Bank signals that the UK government views Fractile as a strategic asset — not merely another hardware startup. Business Secretary Jonathan Reynolds described the company as "a fantastic British success story" in a public statement.
The European context
Fractile is not the only British — or European — bet on AI chips. Dealroom reports that European investments in AI computing hardware surpassed €1.48 billion in disclosed financing during 2026. OLIX (London) raised €270.5 million for inference chips, the Dutch company EUCLYD more than €200 million for custom silicon, Barcelona-based Openchip €115 million in public funding, Valencia's iPronics €107.6 million for optical networking, and startups Callosum, Optalysys, and Tendrils combined €137.8 million.
On the global stage, Tenstorrent (Canada) recently raised $1.5 billion at a $6.2 billion valuation — another inference chip company challenging Nvidia's GPU oligopoly.
The risk
The problem is that Fractile has not yet shipped a single production chip. The company targets data center deployments in 2027. The $7 billion valuation rests entirely on prototypes, unsigned contracts, and one promise: a chip supply agreement of roughly $250 million with Anthropic, reported by Bloomberg but not independently confirmed.
This is the same risk that accompanies all AI hardware startups seeking to challenge Nvidia — the difference being that Fractile has obtained government validation that others have not. The Sovereign AI Fund was created specifically to invest in promising technology ventures and keep industrial computing capability in the United Kingdom.
The question remains whether the bet on memory bandwidth as a differentiating factor will hold when the first chips reach data centers — or whether the industry will continue to prefer Nvidia's established solution, even if it is not the most efficient one.
Sources: TechCrunch, The Next Web, Dealroom
✓ Independent sources cross-checked and verified before publishing
