Fractile Raises $720M Series C for AI Inference Chips with Memory Stacked on Top
$720M, or £540M, for a British chip company that has not shipped a chip yet. Fractile's Series C is led by Thrive Capital, Founders Fund, Lightspeed Venture Partners and Redpoint, and the UK's Sovereign AI Fund is in the round too. The government puts the valuation at "over £5 billion"; FT and Reuters sources say it could be about $7B. TechCrunch reported the confirmation on Wednesday: https://techcrunch.com/2026/11/11/with-720m-round-britains-fractile-gets-a-leg-up-in-the-race-for-ai-chips/
One caveat on timing. The size of the round first surfaced on Oct 29 through Companies House filings, and the filing sums don't all agree (one adds up to £577M). Wednesday's news is the formal confirmation plus the government statement, which says the fund is about "ensuring the U.K. has a greater stake in the hardware that will underpin its future economy."
The technical bet: stack DRAM directly on top of each compute chip, so inference stops being a trade between memory capacity and memory speed, and write its own ML compiler to drive it. For agent workloads that bet makes sense. Long contexts, many parallel sessions and constant KV cache traffic make memory, not raw FLOPs, the bottleneck. CEO Walter Goodwin says the money lets Fractile "ship our first chips and systems next year." Axelera, Etched and OLIX are chasing similar ground.
Read it as a state-backed bet as much as a venture one. A sovereign fund putting money in next to Founders Fund and Thrive says Britain wants a home-grown name in inference silicon instead of renting all of it from Nvidia. $720M buys a first tapeout and systems. Whether it buys customers is a 2027 question.
← Back to all articles
One caveat on timing. The size of the round first surfaced on Oct 29 through Companies House filings, and the filing sums don't all agree (one adds up to £577M). Wednesday's news is the formal confirmation plus the government statement, which says the fund is about "ensuring the U.K. has a greater stake in the hardware that will underpin its future economy."
The technical bet: stack DRAM directly on top of each compute chip, so inference stops being a trade between memory capacity and memory speed, and write its own ML compiler to drive it. For agent workloads that bet makes sense. Long contexts, many parallel sessions and constant KV cache traffic make memory, not raw FLOPs, the bottleneck. CEO Walter Goodwin says the money lets Fractile "ship our first chips and systems next year." Axelera, Etched and OLIX are chasing similar ground.
Read it as a state-backed bet as much as a venture one. A sovereign fund putting money in next to Founders Fund and Thrive says Britain wants a home-grown name in inference silicon instead of renting all of it from Nvidia. $720M buys a first tapeout and systems. Whether it buys customers is a 2027 question.
Comments