A London startup has rocketed to a $3.3 billion valuation in just two years, fueled by a fresh $312 million bet that the future of artificial intelligence demands a radical break from conventional chip design.
The massive Series B round positions OLIX, founded in 2024, as a serious challenger in the race to build hardware specifically for AI inference. This isn’t just another chip company chasing incremental gains. The firm argues the entire industry has hit a wall by using one-size-fits-all processors for the complex, multi-stage work of generating a single AI token.
Backing the raise are major players including Fundomo, Arm, and Hudson River Trading. Angel investors such as Netflix co-founder Reed Hastings also joined the round, while all existing backers doubled down on their commitments. As a result, the company has brought on heavyweights to steer its next phase. Professor Nick McKeown, a Stanford emeritus professor and 2025 Marconi Prize winner, now sits on the board. Matt Briers, formerly of Wise, has signed on as Chief Financial Officer.
OLIX’s core thesis treats the data center like a true factory. Instead of forcing every production step onto a general-purpose chip, its X-1 platform unrolls AI models across a dedicated pipeline of specialized silicon. Each chip focuses on a single part of the process. The design keeps a flexible compute fabric, avoiding a hard-coded architecture, so the system adapts as model designs evolve rapidly.
Perhaps the most distinctive engineering choice sits at the interconnect level. The system uses a “slow and wide” optical network that moves data directly between chips using light, not copper, slashing latency and energy draw. A fully deterministic compiler schedules workloads across racks to make this coherent.
The first product from this architecture, called DX-1, zeroes in on the decode stage where reasoning happens. The company reports that for models with 100 billion parameters, the accelerator delivers over 10,000 tokens per second per user. Critically, DX-1 achieves higher output token throughput per watt than general-purpose chips running large batches. It scales efficiently to models exceeding 10 trillion parameters. The chip relies on fast on-chip SRAM memory, deliberately sidestepping advanced packaging and high-bandwidth memory components that now face severe industrywide shortages.
OLIX aims to ship DX-1 to its first customers by the second half of 2027. To reach that goal, the company will pour this new capital into its custom silicon platform, manufacturing commitments, and a hiring spree across engineering roles in London, Bristol, Austin, Toronto, and San Francisco.
