Two-Year-Old AI Chip Startup Just Raised $312 Million—Here's Why It Matters

Jan de Vries · 2026-08-03

Sometimes a funding round comes along that makes you sit up and take notice. This is one of those moments. OLIX, a London-based AI chip startup, just announced a massive $312 million Series B round at a $3.3 billion valuation. And here's the kicker: the company was only founded in 2024. That's not just fast. That's warp speed in the world of semiconductor hardware, where most startups take years just to get their first chip out the door. ## Who's Betting Big on OLIX? The investor list reads like a who's who of tech royalty. Fundomo, Arm, and Hudson River Trading all participated, alongside some serious angel investors. Reed Hastings, the co-founder of Netflix, is among them. Existing investors also doubled down, which is usually a strong signal that the team is delivering on its promises. The company also made two key hires. Professor Nick McKeown, a Stanford professor emeritus and winner of the 2025 Marconi Prize, is joining the board of directors. And Matt Briers, a former executive at Wise, is coming on board as Chief Financial Officer. These aren't just names on paper—they bring real operational and technical depth. ## The Problem With How AI Chips Work Today Here's the core issue OLIX is trying to solve. Right now, AI data centers run everything on general-purpose chips. Think of it like a factory where every single stage of production uses the same machine, whether it's cutting, welding, or painting. It works, but it's wildly inefficient. OLIX's team puts it this way: "A datacenter is a factory whose product is the token. Producing a single token takes hundreds of operations, each placing different demands on hardware." Their argument is that specialized chips for each stage of the process could unlock a massive leap in performance and cost savings. ## The X-1 Platform: A Different Approach OLIX's answer is the X-1 platform. Instead of one chip doing everything, models are "unrolled" across many chips, creating a production line where each chip focuses on a single part of the model. The key is that the chips stay flexible—they don't hard-code any specific AI model's architecture, which is crucial since AI models are evolving at breakneck speed. The system also uses a novel "slow and wide" optical interconnect. Instead of moving data between chips using copper wires, it uses light. This cuts latency and energy costs dramatically. A fully deterministic compiler schedules workloads across racks, making the whole system work in harmony. ## Meet DX-1: The First Chip in the Lineup DX-1 is OLIX's first chip, designed specifically for the decoding stage—when a model reasons and generates its output. The specs are impressive: - Over 10,000 tokens per second per user for 100B parameter models - Better output token throughput per watt than general-purpose chips at large batch sizes - Scales up to models with 10 trillion parameters or more Perhaps most interestingly, DX-1 uses fast on-chip SRAM memory instead of high-bandwidth memory. It doesn't need advanced packaging either. These are exactly the components that are in shortest supply across the industry right now, so OLIX's design could give it a real edge in scaling production. ## What Happens Next? OLIX plans to use this funding to deliver DX-1 to its first customers by the second half of 2027. The company is also building out its broader custom silicon platform and securing the manufacturing and supply chain commitments needed to scale. If they pull this off, the implications are huge. Cheaper, more efficient AI inference could make frontier AI more accessible and abundant. And it could unlock far more powerful models down the road. OLIX is hiring across silicon, photonics, compiler, and systems engineering in London, Bristol, Austin, Toronto, and San Francisco. If you're in the chip world, this might be a company worth watching—or joining. For now, the $312 million vote of confidence says a lot. This is one to keep an eye on.