> ## Content Index > Fetch the complete content index at: https://nexi.fund/llms.txt > Use this file to discover other available public pages before exploring further. # Fractile's $720M Series C and the UK bid for AI inference silicon - URL: https://nexi.fund/fractile-720m-series-c-inference-chips/ - Published: 2026-11-14T11:30:00.000Z - Updated: 2026-11-14T11:30:00.000Z - Description: Fractile raised $720 million at a reported $7 billion valuation, up from about $1 billion in May, before shipping a production chip. Britain's sovereign funds are buying an option on domestic AI hardware; the 2027 ship date is where the story gets tested. - Author: Nexi.fund Labs - Tags: AI & Infrastructure, #mode-5, #hook-statistic, #track-E, #layout-chronicle-flow On 11 November, Walter Goodwin posted a short update. The company had closed a $720 million Series C. Its first commercial chip would ship next year. The post promised more work ahead and said the new money would fund first shipments. No production silicon exists yet. No disclosed revenue either. Goodwin founded Fractile in 2022 with Yuhang Song. The pitch was narrow and technical. AI models were getting slower to answer because memory bandwidth had become the bottleneck. Fractile would stack DRAM directly on top of logic to fix it. Four years later, that pitch carries a $7 billion price tag. Reuters and the Financial Times both reported the valuation, though the company has not confirmed it. The gap between the valuation and the hardware is the story here: capital is now moving faster than the silicon it is meant to buy. ## The number that reframed the UK chip story Britain has one of the world's most important chip designers in Arm. It has almost no leading-edge manufacturing. The government has spent two years looking for a way to close that gap without building a fab. Fractile became the answer. The Series C included Britain's Sovereign AI Fund and the British Business Bank, alongside Thrive Capital, Founders Fund, Lightspeed and Redpoint. The government described the valuation as "over £5 billion." Public money is buying an option on domestic AI hardware. The option is real; the capacity is not yet. ## From $220M to $720M in six months Fractile's Series B landed in May 2026: $220 million at a valuation near $1 billion, co-led by Founders Fund, Accel and Factorial Funds. Founders Fund returned for the C round. Accel, Conviction, Cocoa, Factorial, Felicis, Gigascale and Kindred also came back. The new money came from Thrive, Lightspeed and Redpoint alongside Founders Fund, with CapitalG, BlackRock, DST Global, Elad Gil, Balderton, SV Angel, Dragoneer, BCV, T. Rowe Price, Georgian, Shopify, 20VC, Nebular, SovAI and Entropy joining. The paperwork tells a slightly larger story. Companies House filings showed £577 million, about $774 million, in Series C shares issued between 9 September and 23 October. That is above the $720 million the company announced. Fractile has not explained the difference. The step-up is unusual even by 2026 standards. A company typically raises a round this size after a product reaches customers. Fractile raised it after publishing a design and a compiler, with revenue still undisclosed. 📊 **By the numbers** $720M Series C, announced 11 November 2026 \~$7B reported valuation, up from \~$1B in May $220M Series B at \~$1B, May 2026 First commercial chip planned for 2027 ## Why inference is the contested layer Training gets the headlines. Inference pays the bills. Every time a model answers a query, it reads its weights from memory. The faster that memory feeds the compute, the faster the answer and the lower the cost per token. Nvidia's HBM stacks sit beside the GPU. Fractile puts DRAM on top of the logic die. The company says this beats HBM on bandwidth and SRAM on capacity. It also ships a compiler, so the chip is sold as a full stack rather than a bare part. The trade-off is familiar to chip designers. HBM offers high bandwidth but limited capacity and a large power budget. SRAM is faster but tiny, which is why it sits close to the compute. Stacking DRAM on logic tries to get capacity and bandwidth in one package, and that is the claim investors are underwriting. It is a manufacturing problem as much as a design one. The compiler matters too: software decides how much of that bandwidth a model actually uses. The design is unproven at scale. It is also why investors paid a premium. If inference becomes the dominant workload, the company that relieves the memory bottleneck owns a toll road. The competitive set is already expensive. Etched raised $700 million at a $21 billion valuation. Olix raised $312 million at $3.3 billion. Cerebras went public in May above $56 billion. Nvidia licensed Groq's technology in December 2025\. Fractile sits between Olix and Etched on valuation with the least silicon shipped of the three, and that ordering is the risk. ## The sovereign-fund question The UK commitment is more than a check. Fractile had about 80 staff in February and had pledged £100 million over three years for UK expansion, including a hardware facility in Bristol. It is hiring in London, Bristol, Palo Alto and Taipei. Public backing lowers the cost of capital and anchors a domestic supply chain. It also ties the company's credibility to an industrial strategy that has changed direction before. For the fund, the round is a visible test of whether public capital can pull private hardware capacity onshore. The Bristol facility matters for a second reason. A domestic hardware site gives the government a physical asset to point to, beyond a valuation mark. It does not yet give Fractile a volume manufacturing line. As we wrote when Qualcomm's HBC silicon reached the lab, passing a benchmark is not the same as volume production (/qualcomm-hbc-inference-silicon-2026/). Fractile has not reached the benchmark stage yet. ## What the 2027 ship date has to prove Bloomberg reported in August that Anthropic agreed an initial order of about $250 million in Fractile chips, with hardware due in 2027\. Neither company has confirmed production volumes. The order is the most concrete demand signal in the story. It is also a promise, not revenue. The arc runs from a 2022 founding to a May 2026 Series B, a November Series C and a 2027 chip launch. Fractile has not named a foundry or packaging partner. Stacked DRAM needs both, and those are scarce inputs across the AI build-out. The $7 billion valuation assumes three things land on schedule. Stacked DRAM has to work in a production package. The compiler has to extract enough performance to beat Nvidia's next generation. Customers have to accept a new vendor for a mission-critical layer. The comparison that matters is Fractile against the version of Nvidia that ships in 2027, when the first Fractile systems reach customers. Any one slips, and the multiple compresses toward the hardware peers that have already shipped. Goodwin's post ended with a version of the same line every chip startup uses: a lot of work ahead. For once, the balance sheet agrees. ## Sources [ UK-backed AI chip startup Fractile hits $7 billion valuation in funding round Reuters confirms the $720 million Series C, the investor list, and the plan to launch a first commercial chip next year. Reuters ](https://live.euronext.com/en/financial-news/uk-backed-ai-chip-startup-fractile-hits-7-billion-valuation-funding-round?ref=nexi.fund) [ With $720M round, Britain's Fractile gets a leg up in the race for AI chips TechCrunch details the DRAM-on-logic architecture, the compiler, and the $5 billion-plus valuation reported by the UK government. TechCrunch ](https://techcrunch.com/2026/11/11/with-720m-round-britains-fractile-gets-a-leg-up-in-the-race-for-ai-chips/?ref=nexi.fund) [ Fractile's $7B valuation: UK chip startup raises $720M six months after $1B Tech Funding News tracks the May Series B, the Companies House filings, and the reported Anthropic order due for 2027 hardware. Tech Funding News ](https://techfundingnews.com/fractile-720m-series-c-7b-valuation-ai-inference-chips/?ref=nexi.fund)