A little-known London semiconductor start-up has jumped into the global spotlight after securing a $220 million funding round, a deal that lifts its valuation to just over $1 billion and underlines rising investor appetite for alternatives to today’s dominant AI hardware stack.

The company, Olix, was founded in 2024 by 25-year-old British entrepreneur James Dacombe. In less than two years, it has attracted roughly $250 million in total financing from a group of well-known venture investors, including Plural, Vertex Ventures, LocalGlobe and Entrepreneurs First. The latest round was led by Hummingbird Ventures, which has previously backed high-profile UK tech names across fintech and consumer internet.

While many start-ups talk about “taking on Nvidia,” Olix is positioning itself slightly differently. Instead of trying to replace GPUs on the heaviest training workloads, the company is aiming at a fast-growing slice of the AI market: inference, the day-to-day work of running trained models in real products. As businesses move from experimentation to large-scale deployment, inference efficiency has become one of the biggest variables in AI operating costs.

That pressure is intensifying as AI systems become more capable. Newer AI agents that can plan, reason, write code, or conduct deeper multi-step research tend to consume far more compute than earlier generations of chatbots. The result is a growing scramble for infrastructure that can deliver high throughput without ballooning power draw, hardware spend, or data-center bottlenecks.

Olix says it is building a “new class of accelerator” designed to push performance while lowering cost, and to do so without being boxed in by the architectural compromises and supply constraints that shape many of today’s AI processors. The start-up describes its approach as an optical digital processor paired with a novel memory and interconnect design, and it says the platform will remain compatible with existing AI models so customers can adopt it without rewriting their entire software stack.

Investors backing the company argue that inference workloads require a more fundamental rethink than simply adapting the same graphics processor lineage that has powered AI’s rise so far. GPUs evolved from video-game rendering into general-purpose parallel compute engines, and they remain extraordinarily powerful. But scaling AI services at mass-market demand has a habit of exposing trade-offs between speed, cost, memory bandwidth, and energy efficiency.

That is why the inference category has become a renewed battleground. For a long time, the AI chip start-up landscape was viewed as brutally difficult and excessively capital-intensive. Yet surging demand for inference capacity has brought the sector back into focus, and high-profile moves by competitors have helped keep investor interest alive. The promise is simple: if a chip can run models cheaper per token and faster per watt, it can reshape the economics of nearly every AI product that reaches consumers or enterprise users.

Olix is also notable for how quietly it has operated. The company was previously known as Flux Computing, and it has said relatively little publicly about product specifications, timelines, or customer commitments. People familiar with its plans say it hopes to deliver first products to customers as soon as next year, though the start-up has not publicly confirmed detailed milestones.

The fundraising is being read in the UK as a rare signal of confidence in domestic hardware ambition. Britain has strengths in chip design and research, but it has historically lagged Silicon Valley in late-stage financing for semiconductor ventures. The contrast is stark: some US-based AI compute firms have raised funding rounds measured in the billions, with valuations that reflect both heavy capital needs and high expectations for market share.

Even so, the UK has an existing footprint. Arm, headquartered in Cambridge, remains one of the world’s most important chip design companies. In AI-specific hardware, the ecosystem has seen turbulence and consolidation, including acquisitions that have shifted ownership overseas. Against that backdrop, backers of Olix hope the company can help the UK claim more value in a booming AI infrastructure market that is still largely shaped by US giants.

Dacombe’s track record adds another layer of intrigue. He is also the chief executive of CoMind, a brain-monitoring start-up he founded as a teenager that has raised significant funding of its own. Supporters say that blend of ambition and execution speed is part of what attracted capital to Olix so early in its life.

Still, the challenge is enormous. In AI compute, success is rarely about chip performance alone. Winning requires software tooling, developer trust, manufacturing partnerships, and proof that systems work reliably at data-center scale. It is also a race against time: models are evolving fast, and infrastructure buyers want roadmaps they can bet their next wave of deployments on.

For now, Olix’s raise is best seen as a bold wager on a specific thesis: that inference has become the next major bottleneck in AI, and that solving it may require a hardware architecture designed from the ground up rather than adapted from the GPU era. Whether the company can convert that thesis into working silicon, real customer adoption, and repeatable performance advantages will determine whether this becomes a defining UK tech story or another cautionary tale in a notoriously unforgiving industry.

The full report was originally published by the Financial Times.