Fractile raises $720 million Series C to build faster, cheaper AI inference chips

Fractile, a British startup that designs AI chips , has closed a $720 million Series C round backed by the UK government's Sovereign ...

Fractile, a British startup that designs AI chips, has closed a $720 million Series C round backed by the UK government's Sovereign AI Fund. The deal puts the company among the best-funded startups building purpose-built AI hardware, and it comes at a time when the cost of running large AI models is a growing worry for the whole industry.

This article covers what was announced, who is behind the round, what Fractile is actually building, and why its approach to memory is the part worth watching.

What Fractile announced

Fractile announced a $720 million Series C. The UK government says the round is worth more than £500 million. Britain's Sovereign AI Fund took part, along with the British Business Bank, the country's state-owned economic development bank.

The government framed the investment as a bet on domestic hardware capability. In its announcement, it said the backing would help make sure the UK has a greater stake in the hardware that will underpin its future economy. The same release describes the aim as reducing dependence on fragile overseas supply chains and keeping more of the jobs and economic value of AI technology inside Britain.

Fractile is based across London and Bristol, and the government says the company plans to expand its presence in Bristol. Business Secretary Jonathan Reynolds called Fractile "a fantastic British success story" and tied the expansion to the government's reindustrialisation plans.

Who invested and at what valuation

The round was led by Thrive Capital, Founders Fund, Lightspeed Venture Partners and Redpoint. The government's own release also lists CapitalG among the backers, alongside Lightspeed and Redpoint, in addition to Sovereign AI and the British Business Bank.

On valuation, the British government said the company is worth "over £5 billion." That converts to roughly $7 billion, a figure that both the Financial Times and a source cited by Reuters reported. Treat the dollar number as reported rather than officially confirmed, since the only figure the government stated is the sterling one.

Fractile also describes itself as a frontier AI company, and the price of the round reflects that framing. Investors are not valuing it like an ordinary chip designer. They are valuing it like a company that sits close to the center of how AI gets built and run.

Where Fractile sits among AI chip startups

Fractile now joins a small group of heavily funded startups chasing purpose-built AI silicon. Axelera, Etched and OLIX are the names most often mentioned alongside it. Etched was reported in early October to be fielding funding offers at a $40 billion valuation, which gives a sense of how much capital is chasing this category.

These chips are one of several routes the industry is exploring to bring down the cost of inference, which means running increasingly demanding AI models in production. Training a model is a large one-time expense. Inference is the cost that repeats every time someone sends a prompt, and it grows as more people and more software agents use AI.

A full-stack approach, not just a chip

Fractile does not want to be seen as a company that only makes chips. The government's release says the company looks at the full stack of technologies that AI rests on, from chips to software to fundamental machine learning research. Part of that stack is a machine learning compiler, the software that translates a model into instructions a chip can run, built to maximize how fast large language models execute on Fractile's hardware.

Fractile CEO Walter Goodwin has described the same philosophy in an October 2026 interview on the No Priors podcast, summarized by Dealroom. According to that summary, the company has about 150 people working across front-end design, physical design, back-end implementation and advanced packaging. The team is deliberately thin in each area but owns the whole silicon process, which lets it iterate in a tight loop instead of handing work to an outside ASIC design house.

That matters because most accelerators are built with the help of such houses. Broadcom is the example Goodwin points to, a company that supplies chip-to-chip connectivity IP and the physical layout that eventually goes to the foundry. Owning more of that chain is slower to set up but can speed up design cycles once it works.

The hardware bet: stacked DRAM

Fractile's main differentiator is probably its memory design. The company's approach now revolves around DRAM, the most common type of RAM, stacked on top of each chip. Fractile hopes this arrangement will get around the usual trade-off between memory bandwidth and memory capacity.

That trade-off needs some explanation. Fast AI chips tend to use one of two kinds of memory. SRAM sits right on the chip and offers extreme bandwidth, but it holds very little data. High-bandwidth memory built from DRAM holds much more, but the supply is tight and the cost is high. A chip that has to choose between the two ends up either fast and small or large and slower.

From SRAM to DRAM

Fractile did not start here. According to the Dealroom summary of Goodwin's interview, the company spent its first two years building an SRAM-based chip, similar in spirit to designs from Groq or Cerebras. By late 2023 it had concluded that growing model sizes and longer context windows made SRAM impossible to scale. It changed direction and began working closely with memory vendors to pull high bandwidth out of cheaper DRAM.

Goodwin also argues that data-center inference economics come down to the cost per gigabyte of memory used. Today's fastest chips pair high bandwidth with low-capacity memory, so tasks that need long context, such as attention over a large conversation history, often have to fall back to a GPU. Fractile wants a design that does both jobs.

Why agentic AI makes this urgent

The memory trade-off has become harder to ignore as agentic AI spreads. Companies want AI systems that carry out long chains of tasks using large models, without slowing to a crawl at each step. Every extra step in a chain means another round of inference, so slow responses compound.

Goodwin has said the real prize is not a snappier chatbot. He compares that to Henry Ford's remark about customers wanting faster horses. The bigger goal is running models with trillions of parameters at many thousands of tokens per second, which would make long-running agents dramatically faster.

Compute is struggling to keep up with this demand, and AI chips are seen as part of the answer. Availability has been limited so far, though. Most of the startups in this field have yet to put production hardware into customers' hands, which is why delivery will matter more than design from here on.

Betting on a constant

AI models change quickly. A new large language model arrives roughly every two weeks, according to Goodwin, and the attention designs in leading open models shift every few weeks. Building hardware for a moving target is risky.

Fractile's answer is to bet on something that does not change. Whatever the architecture of the latest model, each generation wants far more memory bandwidth than the last. Goodwin also argues that a lead of three to six months can decide who wins in deployment, since a single Nvidia system already combines several custom chips and rivals need a real edge to compete.

The UK policy angle

The funding also fits a larger government plan. The announcement came about two weeks after the Budget, where the Chancellor confirmed another £140 million for Sovereign AI. That figure includes £60 million to support the fund's investments and £80 million for its R&D procurement scheme. The Chancellor also said Sovereign AI would expand into physical and industrial AI, including robotics, autonomous systems and advanced manufacturing.

Separately, the UK is pursuing up to £2 billion of investment over the coming years under its AI Compute Roadmap, aimed at building the compute infrastructure the country needs. The Chancellor also raised the British Business Bank's investment capacity by £5 billion, to £30.6 billion.

AI Minister Kanishka Narayan said Britain has too often produced strong scientific ideas only to see the industry and wealth built elsewhere, and called Fractile a case of British ambition in chip design, engineering and AI expertise. Government statements point to Alan Turing and Arm as part of a longer history, and say the aim now is to keep more of the commercial rewards at home.

The government also points out that Sovereign AI typically writes equity checks of around £1 million to £10 million. It does not comment on specific investment amounts, so the size of its stake in Fractile is not public.

What comes next for Fractile

Goodwin acknowledged the road ahead in a LinkedIn post. He wrote that there is still a lot of work ahead, and said the new funding will let the company ship its first chips and systems next year.

That timeline is the thing to watch. Fractile has a large valuation, a notable list of investors and a clear technical thesis. It does not yet have chips in customers' hands. Whether stacked DRAM delivers the bandwidth and capacity Fractile expects will only be settled once real systems run real workloads.

Frequently asked questions

What is Fractile?

Fractile is a British startup, based across London and Bristol, that designs chips for running AI models. It also develops the software that runs on those chips, including a compiler built to speed up large language models.

How much did Fractile raise?

Fractile raised a $720 million Series C. The UK government describes the round as worth more than £500 million.

Who invested in Fractile's Series C?

The round was led by Thrive Capital, Founders Fund, Lightspeed Venture Partners and Redpoint. The UK's Sovereign AI Fund and the British Business Bank also took part, and the government's announcement lists CapitalG among the investors too.

What is Fractile's valuation?

The British government said the valuation is over £5 billion. The Financial Times and a source cited by Reuters reported it as possibly $7 billion.

What makes Fractile's chips different?

Fractile stacks DRAM on top of each chip, hoping to get high memory bandwidth and large capacity at the same time. Most fast AI chips have to give up one for the other. The company also controls much of the design process in house instead of relying on an outside chip design house.

Why did Fractile move away from SRAM?

Its first chip used SRAM, which is very fast but holds little data. According to Goodwin, growing model sizes and longer context lengths made that approach unscalable, so the company shifted to DRAM in late 2023.

What is AI inference, and why does it matter?

Inference is the process of running a trained AI model to produce answers. It is a recurring cost that rises with usage, so cheaper and faster inference hardware can lower the cost of running AI products, especially agents that make many model calls in a row.

When will Fractile ship its first chips?

CEO Walter Goodwin said the new funding will let the company ship its first chips and systems next year.

Why is the UK government investing in Fractile?

The government wants more AI hardware capability built in Britain. It says the investment reduces reliance on overseas supply chains, supports skilled jobs and ensures the UK has a greater stake in the hardware behind its future economy.

Who are Fractile's main competitors?

Startups such as Axelera, Etched and OLIX are in the same category of purpose-built AI chips. Fractile also competes with large players like Nvidia and with custom chips from big technology companies.

References

Fractile Series C announcement on LinkedIn - https://www.linkedin.com/posts/we-are-delighted-to-share-that-fractile-has-ugcPost-7526285169943625747-n5p9/

Walter Goodwin's LinkedIn post - https://www.linkedin.com/feed/update/urn:li:activity:7526304067573182464/

Reuters - https://www.reuters.com/legal/transactional/uk-backed-ai-chip-startup-fractile-hits-7-billion-valuation-funding-round-2026-11-11/

TechCrunch - https://techcrunch.com/2026/11/11/with-720m-round-britains-fractile-gets-a-leg-up-in-the-race-for-ai-chips/

Loaded All Posts Not found any posts VIEW ALL Readmore Reply Cancel reply Delete By Home PAGES POSTS View All RECOMMENDED FOR YOU LABEL ARCHIVE SEARCH ALL POSTS Not found any post match with your request Back Home Sunday Monday Tuesday Wednesday Thursday Friday Saturday Sun Mon Tue Wed Thu Fri Sat January February March April May June July August September October November December Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec just now 1 minute ago $$1$$ minutes ago 1 hour ago $$1$$ hours ago Yesterday $$1$$ days ago $$1$$ weeks ago more than 5 weeks ago Followers Follow THIS PREMIUM CONTENT IS LOCKED STEP 1: Share to a social network STEP 2: Click the link on your social network Copy All Code Select All Code All codes were copied to your clipboard Can not copy the codes / texts, please press [CTRL]+[C] (or CMD+C with Mac) to copy Table of Content