# This British high-school dropout’s startup is now worth $3.3 billion > Source: > Published: 2026-08-03 12:55:06+00:00 [AI](/tag/ai/) A two-year-old British semiconductor startup run by a self-proclaimed high school dropout has raised $312 million in a Series B round – and hopes to take on Nvidia with its proprietary AI inference stack. Olix is betting that it can build better data centre infrastructure by creating specialised chips at each stage of inference, versus running all AI workloads on the same general chip. The chip company, [Olix](https://olix.com/?ref=thestack.technology), was founded by 25-year-old Thiel fellow [James Dacombe](https://www.linkedin.com/in/jamesdacombe/?ref=thestack.technology) in 2024 to build systems for AI inference from the ground up, including chips, networking and lasers. The round includes new investors like Arm, Hudson River Trading and Neflix co-founder Reed Hastings, as well as support from the UK government's venture capital [fund Sovereign AI](https://www.gov.uk/government/news/sovereign-ai-invests-in-uk-startup-reinventing-ai-chips?ref=thestack.technology). ### High school drop out Founder James Dascombe reportedly dropped out of high school at 16 to become a software engineer. In 2022, he was included in [the Thiel fellowship class](https://www.businesswire.com/news/home/20220106005089/en/Thiel-Foundation-Announces-Next-Thiel-Fellow-Class?ref=thestack.technology) while building his first startup CoMind, developing technology that uses light to monitor brain activity. He is still the current CEO of CoMind, which raised $102 million [in October](https://www.comind.io/news/comind-announces-102-5m-funding-to-redefine-clinical-monitoring-of-the-brain?ref=thestack.technology) backed by some of Olix's investors. The Olix technical team includes [Alex Worrall as director of systems](https://www.linkedin.com/in/alexworrall/?ref=thestack.technology), a former engineering manager at Meta, and [Julio Castañeda](https://www.linkedin.com/in/juliocastaneda/?ref=thestack.technology), an alum of autonomous driving startup Luminar Technologies, as the director of mechanical engineering,. ### A new inference approach Olix is building its X-1 system on the premise that in AI datacentres token production shouldn’t rely on one generalist chip. Each stage of a model’s architecture should have its own specialised chip optimised for that particular workload. “Instead of forcing a single chip to do everything, we build specialised silicon and an optical network to connect it, unrolling models across racks like an assembly line,” the company said in a LinkedIn post [announcing the raise](https://olix.com/news/company-raises-series-b?ref=thestack.technology). Splitting model jobs across multiple chips is achievable, the startup says, because of the close rack-scale co-design – the same thesis NVIDIA has bet its [ Vera Rubin](https://developer.nvidia.com/blog/inside-the-nvidia-rubin-platform-six-new-chips-one-ai-supercomputer/?ref=thestack.technology) architecture and [architecture on.](https://siliconangle.com/2026/07/23/co-design-networking-powers-amd-helios-rack-scale-ai-amdadvancingai/?ref=thestack.technology) __AMD’s Helios__A novel network architecture, described as a “'slow and wide' optical interconnect,” uses light to transfer data between chips at low latency. Optimising data centres for faster, lower-cost AI inference means squeezing more potential out of every hardware component. As AI workloads scale, hardware designers are shifting to looking at the rack as a unit of deployment, rather than individual servers or chips. Olix’s funding announcement follows this trend announcing its systems will be delivered as “complete racks for the most demanding inference workloads.” **One size doesn’t fit all ** The first chip the company is developing is a decode accelerator for the model’s reasoning and output stage called the DX-1. It says it hopes to deliver this chip to customers by the second half of 2027. The $312 million in funding announced on Monday will help make this a reality. Olix said it will also use the capital to build “the wider custom silicon platform behind [the chip], together with the manufacturing and supply chain commitments that scaling frontier inference hardware requires.” Alongside the fresh funding, which raised the startup’s valuation to $3.3 billion, Olix announced [Professor Nick McKeown](https://www.linkedin.com/in/nick-mckeown-4902716/?ref=thestack.technology) was joining the company’s board of directors. McKeown is the co-inventor of software-defined networking, another element of the inference rack Olix is innovating on. The company says its system will use “a fully deterministic compiler” to schedule workloads across racks. **Co-design is the future** Frontier labs are all optimising their models to run on specific infrastructure, which is in turn optimised across the hardware and software layer for running AI inference. DeepSeek, for example, is optimised for running on [ NVIDIA chips](https://huggingface.co/nvidia/DeepSeek-V3.1-NVFP4?ref=thestack.technology) and doesn’t perform as well on Google’s TPUs. Founder of Semianalysis Dylan Patel recently extolled the benefits of co-design on the [ Training Data podcast](https://sequoiacap.com/podcast/dylan-patel-of-semianalysis-why-hardware-software-co-design-is-ais-real-100x/?ref=thestack.technology) in June. “The real breakthrough innovation is when you leapfrog a few layers, you co-optimize and co-design them, and now all of a sudden you’ve taken what could have been a 2x here, a 2x here, a 2x here, and instead of being multiplicative to 8x, it’s actually 100x because you’ve optimized it across all three layers,” he said. “That’s what’s really exciting about what you see at the labs, what you see at a company like NVIDIA who’s not co-optimizing on the model layer per se, but a little bit from the model layer all the way downstream to silicon. (...) It’s this co-optimization across many layers of the abstraction stack.”