Fractile raises $720M Series C for AI inference chips

Fractile, the British AI inference chip startup, raised a $720M Series C that the UK government says values it at over £5 billion.
What it means for founders
- More inference hardware is coming, but not before 2027. Fractile joins Axelera, Etched and OLIX among the best-funded startups trying to cut the cost of serving models. If token prices drive your margins, keep compute contracts short enough to move to cheaper capacity once these systems reach clouds.
- One lead investor is betting on long-running agents. Philip Clark of Thrive Capital is most excited by AI agents that work for hours or days at a time, TNW reports. If you build agents, output speed sets how many steps you can afford, so benchmark Fractile systems on your own workloads once they ship.
- Treat the speed claims as unproven. Fractile has no commercial product yet. CEO Walter Goodwin wrote on LinkedIn that the cash will let the company deliver its first chips and systems in 2027, according to TechCrunch.
- UK AI startups have a state co-investor. Sovereign AI says its typical equity checks run around £1 million to £10 million, and the Budget added £60 million to back its investments.
Who led the Fractile Series C
Lightspeed Venture Partners, Thrive Capital, Redpoint and Founders Fund led the round, according to TechCrunch and TNW. Other backers named by TNW include BlackRock, CapitalG, T. Rowe Price, DST Global, Balderton, Shopify Ventures and Bain Capital Ventures. The government release names CapitalG, Lightspeed and Redpoint and describes the round as worth more than £500 million. Silicon UK converts the $720 million to £542 million.
Both the Financial Times and a source speaking to Reuters put the valuation at $7 billion, TechCrunch says. In May, Fractile raised about $220 million at roughly $1 billion.
How the Fractile AI inference chip works
TNW reports that over 25 years, processor performance has risen about a millionfold while memory bandwidth grew around 40 times. Goodwin told the FT that leading models are stuck "drinking through a straw."
The current Fractile design stacks custom DRAM and logic in three dimensions. The company says it can deliver much more bandwidth than the high-bandwidth memory (HBM) many rivals rely on, while holding enough capacity for big models working over long contexts. Goodwin claims the chips could run inference up to 10 times faster. By skipping HBM, Fractile also hopes to sidestep the demand crunch at SK Hynix, Micron and other memory makers. A machine learning compiler sits on top to speed up how language models run on the silicon.
A crowded field of AI chip startups
Last year Nvidia paid to license Groq's inference technology and hire its leaders, and in March it released its own inference chip. Etched, MatX and OLIX are building alternatives, and Tenstorrent raised $1.5 billion in October. Fractile also has sites in Taiwan and the US.
What we don't know yet
Reports on Anthropic differ. In August, Bloomberg, citing people familiar with the matter, reported an initial agreement for Anthropic to buy roughly $250 million of Fractile chips, TNW reported. The FT now describes Anthropic as interested in becoming an early customer. Both companies declined to comment to Bloomberg at the time.
The government does not disclose how much Sovereign AI put in, and Fractile has not posted a release on its website. The government gives the round's size only as more than £500 million.
The story
Fractile, the London and Bristol startup designing AI inference chips, has closed a $720 million Series C, TechCrunch reported on November 11. The UK government, which invested through the British Business Bank and its Sovereign AI fund, says the round values Fractile at over £5 billion.
Sources
Primary sources
Reporting
More in Funding & Business
- TSMC to invest up to $1B in Anthropic and OpenAITSMC's board approved up to $500M each for Anthropic shares at the IPO price and OpenAI preferred stock.Ars Technica · 6d ago

- OpenAI revenue tops $10B in a quarter, Bloomberg says

For founders: Discounts helped drive the jump. TechCrunch lists tighter pricing and discounts next to better models as reasons for the rebound, and Techzine points to steep…
TechCrunch · 7d ago - Nvidia board weighs ousting Jensen Huang

For founders: Nvidia's product plans look set for now. Futurum's Shay Boloor told Yahoo Finance that Blackwell demand and the Rubin roadmap were firmly established long…
TechCrunch · 6d ago - Anthropic sets IPO range at $70 to $78 a share

For founders: Pricing will be read against quarterly results. With $900.9 billion committed to compute, every change to Claude plans, API rates and rate limits will be…
TechCrunch · 8d ago - Anthropic said to push ahead with IPO next week

For founders: The range sets a public price on frontier AI revenue. Expect investors to treat whatever number Anthropic and its banks settle on as a reference point when…
TechCrunch · 8d ago