London-based AI inference chip startup OLIX raises $312 million in funding at £2 billion valuation to scale frontier inference silicon.

OLIX, a two-year-old London-based AI chip startup founded by James Dacombe, has announced a $312 million Series B financing round at a $3.3 billion valuation. The round includes Fundomo, Arm, and Hudson River Trading, alongside angel investors including Netflix co-founder Reed Hastings. Existing investors Hummingbird Ventures, Crane, Plural, Creandum, Phoenix Court, and Transition all increased their commitments.

OLIX designs every part of its systems—the chips, the lasers, and the network that connects them. Its systems are delivered as complete racks for the most demanding inference workloads.

The company believes the industry's approach to building AI inference hardware is reaching its efficiency limits. OLIX frames a datacenter as a factory whose product is the token; producing a single token takes hundreds of operations, each placing different demands on hardware. While any traditional factory would assign each stage a machine built for it, the token factory runs every stage on the same general-purpose chip. Each new chip generation has improved by pushing single-chip specifications higher—an even better generalist, but never a specialist. OLIX's systems are built on the belief that using specialized chips for each stage of the token production process will unlock a step change in AI performance and cost.

In OLIX's X-1 platform, models are fully unrolled across a large number of chips, creating a production line that allows each chip to focus on a single part of the model. As AI model architectures rapidly evolve, each chip retains a flexible compute fabric and does not hard-code or bake in a particular model's architecture.

The system is built around a novel "slow and wide" optical interconnect that moves data directly between chips using light instead of copper, at ultra-low latency and energy cost. This is made possible by rack-scale codesign across every part of the link. Workloads are scheduled across racks by a fully deterministic compiler. OLIX believes this will make existing frontier AI more affordable and abundant and, more importantly, will unlock the deployment of far more powerful models in the future.

The first chip OLIX is building within the X-1 platform is DX-1, its decode accelerator, designed for the stage at which a model reasons and generates its output. For 100-billion-parameter models, DX-1 can achieve pareto-optimal inference performance, delivering over 10,000 tokens per second per user at higher output token throughput per watt than general-purpose chips running large batch sizes. The architecture scales performantly to models of 10 trillion parameters and above, thanks to OLIX's multi-rack scale-up domain architecture.

DX-1 holds a model in fast on-chip memory (SRAM) for higher energy efficiency and lower latency. The design uses no advanced packaging and no high-bandwidth memory—the components the industry is in shortest supply of—and is designed to scale volumes despite supply chain shortages across the industry.

The financing funds the path to deliver DX-1 to OLIX's first customers by H2 2027 and the build-out of the wider custom silicon platform behind it, together with the manufacturing and supply chain commitments that scaling frontier inference hardware requires.

OLIX is hiring across silicon, photonics, compiler, and systems engineering in London, Bristol, Austin, Toronto, and San Francisco.