⚡️ DX-1 on board

Olix 🇬🇧 employs 140 people across London, Bristol, Toronto, Austin and San Francisco. Two continents, five cities, one roadmap: ship the DX-1, its first photonic inference accelerator, before the end of 2027.

The round that just landed, a €270m Series B at a €2.9bn valuation, triples the numbers Olix was showing six months ago. Nvidia builds general-purpose silicon. Olix is betting on the opposite: a chip that does one thing, decoding tokens as they leave the model, and does it fast. The DX-1 claims over 10,000 tokens per second per user on a 100bn-parameter model, at lower power draw than general-purpose silicon running the same workload. Cerebras and Groq are chasing the same shift with different architectures; neither has bet on photonics, and whose approach ages better is still an open question. James Dacombe, 25, runs Olix alongside a second company, CoMind, working on brain-machine interfaces.

The round is led by Fundomo, a New York fund, with Hudson River Trading and angel investor Reed Hastings on the US side. On the European side, Arm 🇬🇧 has signed off on the technical case, while Creandum 🇸🇪 and the UK’s 🇬🇧 sovereign wealth fund topped up existing stakes. The round is being run out of New York. The product itself is still taking shape in London and Bristol.

Dealroom puts Europe’s share of global venture capital flowing into AI chips at 6% between 2022 and 2025. Arago 🇫🇷, VSORA 🇫🇷 and Fractile 🇬🇧 are digging the same photonic niche out of Paris and Cambridge regardless. McKinsey reckons deep tech could generate €870bn of value in Europe by 2030, provided growth capital shows up. One floor up, in infrastructure, Nscale 🇬🇧 just closed €2.9bn in London, the largest European raise of the year: the money exists to build data centres, not yet to design the chips running inside them. On this specific measure, the gap sits in the size of the cheques, not the quality of the research.

A continent that files the patents, trains the engineers and licenses its intellectual property to whoever can deploy it at global scale is playing the role of a luxury technology subcontractor, not unlike Swiss watchmaking before it went global. Not everyone will accept that comparison: the DX-1 itself is still being engineered on this side of the Atlantic, IP and product both. But whoever writes the cheques decides which layer of the stack gets kept in the long run. Who leads this particular round is not the story. It’s the production capacity the DX-1 will need by 2027, and no European fund can finance that alone right now.

The next round won’t settle the industrial question on its own. It will hand out the clearest signal yet: who ends up financing DX-1’s production, and where.