The 25-Year-Old Who Dropped Out of Sixth Form and Built a $3.3B AI Chip Company

James Dacombe left sixth form after two days. He was 16.

Nine years later, at 25, Dacombe is Britain's newest billionaire. Olix just raised $312 million at a $3.3 billion valuation, tripling in six months.

I have been watching the custom silicon race for two years now, and I have never seen a chip startup move this fast. The funding round, announced August 3, tripled Olix's valuation in six months.

It was led by New York-based Fundomo. Arm invested. Netflix co-founder Reed Hastings put in money.

Hudson River Trading joined. Sovereign AI, the UK government's vehicle for backing promising AI companies, wrote a check too.

"We are on track to be the fastest company to go from founding to full chip tape out of any company in this space," Dacombe said.

The company builds inference chips, the specialized silicon that runs trained AI models at scale. Its approach ditches copper interconnects for photonics, moving data between chips with beams of light.

Olix calls it a "slow and wide" optical fabric. The first silicon, a decode accelerator called DX-1, targets models in the 100-billion-parameter range.

The numbers Olix is claiming are aggressive. For a 100B parameter model, DX-1 promises more than 10,000 tokens per second per user. That's simultaneous high throughput and low latency, a combination that general-purpose GPUs struggle to deliver.

The architecture has no high-bandwidth memory and no advanced packaging. No HBM.

That design choice is deliberate. Olix keeps model weights in fast on-chip SRAM, avoiding the supply chain bottlenecks that have made HBM one of the scarcest components in the semiconductor industry. The chip can scale to models above 10 trillion parameters by spreading across multiple racks.

"I'd be surprised if the architecture scales perfectly on the first silicon," I found myself thinking while reading the technical summary. Chip startups always print ambitious numbers. The interesting question is not whether they hit 10k tokens per second on day one, but whether they get close enough to change procurement conversations inside the hyperscalers.

Olix plans to finalize the DX-1 design this year and ship first customer racks in the second half of 2027. The company now employs more than 140 people across London, Bristol, Toronto, Austin, and San Francisco. It pays employees a £2,000 monthly bonus to live within 20 minutes of its King's Cross office.

Dacombe started his first company at 17. CoMind, a medical technology startup based on non-invasive brain monitoring, was inspired by watching relatives affected by strokes and dementia. He built it with a $250,000 grant from Peter Thiel's fellowship program, which pays young entrepreneurs to skip university and build companies instead.

"The velocity and culture of delivering when you say you will deliver is really key," Dacombe said.

His mother Jane, a freshwater ecologist at the Environment Agency, posted on LinkedIn after the round was announced: "An alchemy of your vision and drive together with an amazing team. So very proud of you."

Alongside the funding, Olix appointed Nick McKeown, the Stanford professor who co-invented software-defined networking, to its board. It also brought in Nick Briers, the former CFO who took Wise public on the London Stock Exchange in a landmark direct listing.

Joséphine Kant, head of ventures at Sovereign AI, said she left her first meeting with Dacombe "convinced he'd be one of the most important founders of his generation." AI minister Kanishka Narayan said Olix was "exactly the kind of ambitious company we want to back through Sovereign AI."

The context for all of this is Nvidia, which holds roughly 80% of the AI chip market and is valued at just under $5 trillion. Olix is betting that the future of inference looks more like a factory than a Swiss Army knife, with specialized machines for each stage of token production and fast, cheap movement between them.

It's a bet worth watching. The DX-1 tape-out, expected later this year, will be the first real data point.

Until then, a $3.3 billion valuation for a pre-revenue chip startup with a 25-year-old founder is either the canniest trade in European venture capital or the most expensive. The tape-out will tell us which.

Source