OLIX Raises $312 Million to Run AI on Light, Not Electricity

Almost every story about the cost of AI ends up at the same two walls: there are not enough chips, and there is not enough power to run the ones we have. A British startup called OLIX has just raised 312 million dollars on a bet that the way around both is to stop pushing electrons through silicon and start pushing light.

The Series B values the two-year-old company at 3.3 billion dollars and is, according to Data Center Dynamics, the largest semiconductor venture round a European company has ever closed. The New York firm Fundomo led it. The rest of the table is telling: the chip designer Arm, the trading house Hudson River Trading, Netflix co-founder Reed Hastings, and the UK government's own Sovereign AI Fund all put money in.

What an optical chip actually does

OLIX makes what it calls Optical Tensor Processing Units. The heavy arithmetic behind a neural network, multiplying long lists of numbers over and over, is done by beams of light passing through the chip rather than by transistors switching on and off. Light runs fast, generates little heat, and can carry many signals at once. In theory that attacks the energy problem at its root instead of building bigger cooling systems around it.

The company's pitch, laid out by Converge Digest, is a rack-scale design that links more than 100,000 chips with optical connections and, notably, drops the high-bandwidth memory that sits on today's GPUs. HBM is one of the tightest supply chokepoints in the industry, so a credible way to do without it matters. This is the same bottleneck a new memory standard is trying to widen from the other direction.

The catch: it does not exist yet

Here is the sober part. OLIX's first product, an inference chip called DX-1, is not due until the second half of 2027. The money buys the runway to build it. Optical computing has been ten years away for a long time, and plenty of well-funded photonics ventures have run into the same problem, that shipping a working, programmable, manufacturable chip is much harder than showing the physics works on a bench. The appointment of the Stanford networking pioneer Nick McKeown to the board is a signal that OLIX knows this and is buying credibility along with talent.

Why the state money is the real story

What stands out is not the physics but the flag. A national government backing a specific chip startup is a bet that sovereign computing capacity is worth owning, the same instinct behind AMD buying its way into etched-silicon inference and Anthropic building its own chip team. Nobody wants to depend entirely on a single American supplier for the hardware that now underpins their economy.

Whether OLIX ships is a 2027 question. That a European government helped write a nine-figure cheque to find out is an answer about 2026, and about how nervous everyone has become about who controls the picks and shovels.