New AI‑accelerator startups fundraise

European AI chip startups are lining up nine-figure fundraises as investors hunt for alternatives to Nvidia in the next phase of artificial intelligence computing. (cnbc.com) Dutch startup Euclyd told CNBC on April 17 that it is discussing a round of at least €100 million, or about $118 million. Founder Bernardo Kastrup said the company has raised less than €10 million so far and wants new capital to scale and reach first customers. (cnbc.com) In London, Fractile has been in talks to raise more than $200 million at a $1 billion valuation, the Financial Times reported on March 31. The company was founded in 2022, previously raised about $15 million in seed funding, and counts former Intel chief executive Pat Gelsinger and the NATO Innovation Fund among its backers. (ft.com) These companies are aiming at AI inference, the step where a trained model answers a prompt or serves a user request, rather than the earlier training phase. CNBC reported that investors are shifting attention from the chips used to build giant models toward the hardware needed to run them cheaply and at scale. (cnbc.com) That shift has pulled in more capital across Europe. Crunchbase said European venture funding reached $17.6 billion in the first quarter of 2026, up nearly 30% from a year earlier, with artificial intelligence taking more than half of the continent’s funding for the quarter. (crunchbase.com) The money is also concentrating in chip startups. CNBC reported that investors have already put more than $200 million into Axelera AI and the U.K.’s Olix in 2026, while Axelera said on February 24 that its own latest financing totaled more than $250 million. (cnbc.com) (axelera.ai) The technical pitch is similar across many of these startups: move data around less, use memory differently, and cut the power bill for every answer an AI model produces. Fractile said its design uses in-memory compute, while the Financial Times reported its chips rely on static random access memory, or SRAM, rather than Nvidia’s standard graphics processor approach. (fractile.ai) (ft.com) Euclyd is making a similar efficiency argument. CNBC reported that the startup says its chip system can deliver 100 times higher power efficiency for inference than Nvidia’s Vera Rubin generation, though Nvidia did not respond to CNBC’s request for comment. (cnbc.com) Europe’s pitch is not only technical. CNBC said Euclyd is backed by former ASML chief executive Peter Wennink, and Fractile investor Patrick Schneider-Sikorsky of the NATO Innovation Fund pointed to U.S. export controls, Taiwan Semiconductor Manufacturing Company concentration risk, and demand for “sovereign” compute as forces pushing money toward local chipmakers. (cnbc.com 1) (cnbc.com 2) The harder part comes after the fundraising. CNBC said Europe still has a thinner funding base and a smaller manufacturing footprint than the United States and Asia, and the Financial Times noted that Fractile’s round was not finalized and terms could still change. (cnbc.com) (ft.com) For buyers of AI hardware, the immediate result is a wider field. Instead of comparing only Nvidia graphics processing units, cloud providers and model developers are being asked to weigh new chips on speed, cost, power use, software support, and whether the startups raising nine-figure rounds can actually ship. (cnbc.com)