$312 Million for Olix: UK AI Chip Startup Valued at $3.3 Billion

Two years old and already worth $3.3 billion — that's the trajectory Olix is on after closing a $312 million Series B. The London company, which designs its own chips, lasers and networking gear for AI inference, only crossed the unicorn line in February with a $220 million Series A, one of the largest ever raised by a UK tech startup at that stage. Six months on, the valuation has more than tripled.

The investor list reads like a who's-who of people betting on the inference boom. Fundomo, a New York deep tech firm already in the February round, is said to have led again. Arm and high-frequency trading firm Hudson River Trading joined in, alongside angel money from Netflix co-founder Reed Hastings. Existing backers including Hummingbird Ventures, Crane, Plural, Creandum, Phoenix Court and Transition all topped up. And then there's a name Olix didn't mention in its own announcement: Sovereign AI, the UK state venture fund, which the government confirmed separately as a participant. It's the fifth equity investment the fund has made since launching, out of eleven startups it supports overall. AI Minister Kanishka Narayan framed it as a leverage play — countries that build chips build influence. Sovereign AI's Joséphine Kant even name-checked founder James Dacombe's holding company, Bletchley Industries, a nod to the codebreakers of Bletchley Park.

Dacombe founded Olix in 2024 at age 23, having previously built neurotech company CoMind. His pitch is that the industry's go-to approach to inference hardware — running every kind of computation through the same general-purpose chip — is hitting a wall. Olix compares a data centre to a factory that manufactures tokens, and argues no sane factory would run every step of production on one machine. Its platform, X-1, spreads a model across many chips, each one handling just a slice of it, connected by what the company calls a slow and wide optical interconnect that moves data with light rather than copper. A deterministic compiler handles the traffic direction.

The first chip out of that platform, DX-1, targets the decode phase — where a model actually generates output. Olix claims more than 10,000 tokens per second per user for 100-billion-parameter models, with better throughput per watt than general-purpose chips running large batches, and says the design should scale up to models of 10 trillion parameters and beyond. None of that has been independently verified yet. What's notable is the memory story: DX-1 is built to run entirely on fast on-chip SRAM, skipping advanced packaging and high-bandwidth memory altogether — the exact components the whole chip industry is currently scrambling to secure.

Olix is also stacking its leadership bench. Nick McKeown, the Stanford professor emeritus behind software-defined networking, OpenFlow and P4, and a 2025 Marconi Prize winner, is joining the board; he's previously sold startups Nicira and Barefoot Networks to VMware and Intel. Matt Briers, who spent nine years as Wise's CFO and steered its 2021 direct listing — the first by a tech company on the London Stock Exchange — takes over as Olix's finance chief.

All of this lands in a crowded, well-funded field. Dealroom pegs global AI chip startup funding at roughly $8.3 billion for 2026 alone, driven by the shift from training models to actually running them at scale for agents and reasoning tasks. Nvidia bought Groq's assets for $20 billion in December and then put another $4 billion into photonics firms. Cerebras just raised $1 billion at a $23 billion valuation, d-Matrix pulled in $275 million, and Etched sits around a $5 billion valuation. Olix says the new cash will get DX-1 to its first customers by the second half of 2027 and fund the manufacturing and supply chain work needed to build out the rest of the platform, with hiring underway across London, Bristol, Austin, Toronto and San Francisco.