Arm and Netflix co-founder Reed Hastings are among investors to have lent backing to Olix, an AI infrastructure startup that counts one of the pioneers of software-defined networking on its board.
The London-based startup was founded in 2024 by Thiel Foundation fellow James Dacombe. Valued at $3.3 billion, it aims to build “the infrastructure for frontier AI inference” with specialized chips built around an optical interconnect it claims provides ultra-low-latency performance.
Olix raised $312 million in a Series B round that included venture capital giants Fundomo and Hudson River, with the firm adding that existing investors increased their commitments. The funds will be used to ensure deliveries are shipped to Olix’s first customers by H2 2027, as well as to build out its wider custom silicon platform and to hire staff across London, Bristol, Austin, Toronto, and San Francisco.
Alongside the cash raise, Olix gained a boon with the addition of Nick McKeown to its board of directors. Longtime readers will know McKeown as one of the minds that helped coin SDN (software-defined networking) following work at Stanford University. He’d later found Nicira Networks which VMware snapped up in 2012, and Barefoot Networks, which Intel acquired in 2019.
In a LinkedIn post, McKeown said it was a combination of confidence in Dacombe and network needs to boost inference performance that caught his attention.
“Inference is bounded by moving data, not by arithmetic... Olix’s approach stands out because they refuse to treat the chip, the interconnect, and the compiler as separate problems,” he wrote. “Designing them together, with light doing what copper no longer can, is how you get a step change instead of just another increment. It was their creative full systems thinking that got me involved.”
Olix’s so-called “slow and wide” optical interconnect moves data directly between chips using light instead of copper. While several firms are already making headway in the optical interconnect space, the British startup is looking at the idea at the rack-scale level “across every part of the link.” Each workload is scheduled across racks by a fully deterministic compiler, an approach not too dissimilar from that seen in Nvidia’s Groq 3 language processing unit (LPU).
Like the Nvidia Groq chip or Cerebras’ wafer-scale lines, Olix is backing static random-access memory (SRAM) – fast on-chip memory – to provide lower latency when running an AI model. The startup said its choice means it can potentially scale given the lack of supply-chain-constrained high-bandwidth memory.
“Olix's mission is to build the infrastructure for frontier AI. Instead of forcing a single chip to do everything, we build specialised silicon and an optical network to connect it, unrolling models across racks like an assembly line,” Dacombe wrote in a LinkedIn post. “A different answer to the industry's supply chain constraints, built on components available at scale today.”