Olix Computing is a London-based semiconductor startup designing a fundamentally different AI accelerator: the Optical Tensor Processing Unit (OTPU). Instead of the GPU-plus-high-bandwidth-memory (HBM) architecture that dominates AI inference today, Olix combines on-chip SRAM with photonics — using light rather than purely electronic signaling for tensor operations. The company's bet is that an SRAM-and-photonics design can surpass HBM-based architectures on throughput-per-watt and total cost of ownership, attacking two of the biggest pain points in scaling AI inference: energy and cost.
The company was founded in 2024 by James Dacombe, who at 25 is one of the youngest founders leading a UK unicorn in one of tech's most capital-intensive sectors. Dacombe is also CEO of CoMind, a brain-monitoring startup he started as a teenager that has raised around $100 million. His background spans neurotechnology and deep hardware, and Olix represents an ambitious move into frontier AI silicon at a moment when the industry is hungry for alternatives to Nvidia's GPU stack.
In February 2026, Olix raised a $220 million Series A at a valuation exceeding $1 billion, bringing total funding to roughly $250 million since its 2024 launch. The round was led by Hummingbird Ventures — whose portfolio includes Kraken, Revolut, and Deliveroo — with participation from Plural, LocalGlobe, and Entrepreneurs First. The size of an early Series A reflects the enormous capital required to design and tape out novel chips, as well as investor appetite for credible challengers in AI compute.
Photonic computing has long promised dramatic efficiency gains for AI by moving data with light, but commercialising it has proven difficult. Olix joins a small but growing cohort of photonic and optical AI chip companies racing to prove the approach can work at production scale. Its specific angle — eliminating reliance on expensive, supply-constrained HBM by leaning on SRAM and photonics — differentiates it from both conventional accelerators and other optical startups.
Olix plans to ship its first products to customers in 2027, meaning its near-term value rests on technical milestones and tape-outs rather than revenue. If it delivers, the payoff could be substantial: a more energy- and cost-efficient inference chip would be highly attractive to hyperscalers and AI labs straining against power and HBM constraints. The risks are equally clear — photonics is hard, timelines in silicon slip, and incumbents are formidable.