Fractile raises $720 million for AI chips it has yet to ship, four years after founding

Fractile, a London- and Bristol-based maker of AI inference chips founded in 2022, has raised $720 million in a Series C round led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, the company said on its LinkedIn page on Wednesday, November 11. The UK government said the round values Fractile at over £5 billion.

A seed round built on simulations

Fractile was founded "in the summer of 2022," according to its website, on the bet that the most capable AI systems would be limited by the time they take to produce useful outputs. Its founder and CEO, Walter Goodwin, was a 28-year-old PhD graduate of the University of Oxford's Robotics Institute when the company came out of stealth in July 2024, Fortune reported at the time.

That first round was a $15 million seed led by Oxford Science Enterprises, with Kindred Capital and NATO's innovation fund, according to Fortune. The company then employed 14 people and had "only tested its designs in computer simulations," the magazine reported, while Goodwin said the simulations showed its chips could be "100 times faster and 10 times cheaper" than Nvidia's graphics processors for running AI models.

From one idea about memory to stacked DRAM

In 2024 Goodwin described a design that stored data "directly next to the transistors" doing the arithmetic, according to Fortune. The approach the company describes now is different: dynamic random-access memory (DRAM) stacked in three dimensions on top of the logic. Fractile's post says this "co-design of DRAM and logic stacked in 3D" gives better bandwidth and speed than high-bandwidth memory, the kind used beside most AI accelerators. TechCrunch described the change as a pivot.

Lightspeed, which co-led the new round, said in its investment note that "AI chip efforts often hand backend design to Broadcom or another outside partner; Fractile owns its entire stack."

Two large rounds six months apart

The company announced a $220 million round on May 13, 2026, in a post by Goodwin. The new round is about 3.3 times that size, by FinanceInTime's calculation. Beyond the four leads, Fractile listed CapitalG, DST Global, Elad Gil, BlackRock and Bain Capital Ventures among participants.

The UK government's Sovereign AI Venture Fund and the British Business Bank also invested, in a round the government put at "over £500 million." TechCrunch noted that the Financial Times and Reuters reported a valuation of $7 billion. In May, Fractile said it was hiring in London, Bristol, Palo Alto and the Bay Area, and Taipei.

In the founder's words

Goodwin's May essay set out the problem as the company sees it: models already produce "up to 100 million tokens" on hard problems, and "at the ~40 tokens per second or so at which these models tend to run on existing chips, a single output of this length takes a month to complete."

"Fractile's journey has only just begun, and the most important work lies ahead," he wrote then. In a LinkedIn post about the new round, Goodwin wrote that the funding will let the startup "ship [its] first chips and systems next year," TechCrunch reported.

The company's stated plan

Business Secretary Jonathan Reynolds said Fractile has "plans to expand their presence in Bristol." The company has not published revenue, customer names or a date for its first chips beyond "next year."

A round that pays for first silicon, not proof

We think Fractile's $720 million is a milestone, not evidence that the design works: in July 2024 it had only simulations, and the company still has no chips in customers' hands. What the funding does show is that investors and the UK state are willing to back a full-stack chip team for a second large round inside six months.

The opportunity is real if the stacked-memory design delivers the speed the company describes, because inference speed is the constraint that both Goodwin's essay and Lightspeed's note describe. The threats are the familiar ones for chip startups: first silicon that slips, manufacturing partners with limited capacity, and a small pool of buyers able to commit to new hardware at scale. We would watch whether the first chips ship in 2027 as Goodwin said and whether a named customer follows; our view would change if a customer commits publicly before the chips ship.

The view in this final section is FinanceInTime's editorial opinion, for information only. It is not investment, tax, legal or other advice, and it does not take your circumstances into account. Forecasts are uncertain and past results do not guarantee future ones.