How London's OLIX AI Chip Startup Aims to Shatter the Interconnect Bottleneck

The fundamental bottleneck in modern artificial intelligence is no longer compute; it is communication. As neural networks scale, the copper wires linking chips together have become a physical bottleneck, choking throughput and inflating power bills. By raising a $312 million Series B at a $3.3 billion valuation, London-based OLIX, the high-flying AI chip startup, is betting that the path to sovereign hardware independence lies in replacing those copper wires with light.

The Multi-Billion Dollar Fight Against the Silicon Wall

Founded in 2024 by 25-year-old serial entrepreneur James Dacombe, OLIX has transitioned from stealthy hardware underdog to one of Europe’s most valuable deep-tech players. The new $312 million injection, coming hot on the heels of a $220 million Series A earlier in the year, is backed by heavyweights including Fundomo, semiconductor giant Arm, high-frequency trading firm Hudson River Trading, and Netflix co-founder Reed Hastings. Notably, the round also features direct backing from the UK Sovereign AI fund, highlighting the geopolitical stakes undergirding the hardware stack.

To understand why a company founded less than two years ago commands a multi-billion dollar valuation, one must look at the physics of the modern data center. Today's frontier large language models (LLMs) cannot fit onto a single graphics processing unit (GPU). Instead, they must be sliced and distributed across thousands of chips, all of which must continuously synchronize parameters. At these scales, conventional copper interconnects consume too much power and generate excessive heat, degrading performance.

The OLIX AI chip startup is addressing this bottleneck by engineering a modular, full-stack inference architecture. Rather than building just another chip to compete with Nvidia’s H100 or Blackwell architectures, OLIX is designing custom silicon, co-packaged silicon photonics (lasers), and specialized optical networking. The goal is to build entire, self-contained AI inference racks where data moves between processors at the speed of light.

Why Silicon Photonics is the Holy Grail of AI Inference

Silicon photonics—the integration of microscopic lasers and optical waveguides directly onto silicon dies—has been a promised land for semiconductor engineers for over a decade. But while industry pioneers like Lightmatter, Celestial AI, and Ayar Labs have focused on selling individual components or optical bridges to existing chipmakers, OLIX is taking a vertically integrated, system-level approach.

To solve the supply chain crisis and the energy wall, you cannot just tape out a better accelerator. You have to design the chip, the laser, the optics, and the network topology as a single, coherent machine.

James Dacombe, CEO and Founder of OLIX

By controlling the entire stack from the specialized silicon up to the rack-level networking, OLIX claims it can deliver AI inference that is orders of magnitude faster, cheaper, and more energy-efficient than traditional GPU clusters. This focus on inference is strategic. While model training is a centralized, bursty process dominated by hyperscalers, inference is an ongoing operational cost that scales with user adoption. In a world where AI agents execute millions of background tasks daily, optimizing cost-per-token is the only metric that matters.

Sovereign AI and the Geopolitics of UK Deep Tech

The participation of the UK Sovereign AI fund in this round is a loud signal. Historically, Britain has been excellent at spawning world-class deep tech companies—only to watch them flee to Silicon Valley or be acquired by foreign conglomerates (a fate Arm itself narrowly navigated through its SoftBank era). Dacombe, who previously founded the London-based neurotech startup CoMind, represents a new cohort of European founders determined to scale global infrastructure companies from London.

To support this rapid scaling, OLIX has assembled an elite tier of leadership. Networking legend Professor Nick McKeown, who co-founded Barefoot Networks (acquired by Intel) and revolutionized software-defined networking, has joined the OLIX board of directors. Alongside him, former Wise chief financial officer Matt Briers has been appointed as CFO to manage the capital-intensive deployment of the company's first commercial wafer runs and manufacturing lines.

For the UK government, backing OLIX is about strategic autonomy. If the next generation of computing is optical, having a domestic champion that controls the intellectual property for co-packaged optics, laser integration, and modular rack architectures ensures that Europe remains a player, rather than just a customer, in the physical AI landscape.

The Execution Challenge: Scaling Hardware is Brutal

Despite the immense capital and talent OLIX has accumulated, the path ahead is fraught with execution risk. Upending established semiconductor supply chains is notoriously difficult. Giants like Intel, Broadcom, and Nvidia (via its NVLink optical roadmaps) are investing billions into their own silicon photonics portfolios. OLIX must prove not only that its technology works in a lab, but that it can be manufactured reliably at scale using existing foundry processes.

Furthermore, building custom hardware requires building a developer ecosystem. Nvidia's moat is not just its hardware, but CUDA—the software layer that engineers have spent fifteen years optimizing. OLIX will need to deliver a highly optimized software compiler that allows developers to run models seamlessly on their optical racks without rewriting their codebases from scratch.

A Light-Speed Paradigm Shift

The OLIX AI chip startup's $3.3 billion valuation represents more than just investor enthusiasm; it is a structural bet on the post-copper era of computing. If Dacombe and his team can successfully deploy modular, laser-linked inference racks, they will rewrite the economics of serving AI models at scale. For founders, enterprise buyers, and sovereign states, the message is clear: the future of AI will not be won just by making transistors smaller, but by making data travel faster.