ChannelLife UK - Industry insider news for technology resellers
United Kingdom
ARIA tests Lumai optical AI servers in data centre

ARIA tests Lumai optical AI servers in data centre

Tue, 17th Nov 2026 (Today)
Joseph Gabriel Lagonsin
JOSEPH GABRIEL LAGONSIN News Editor

ARIA will deploy Lumai's optical compute technology in its Scaling Inference Lab, in what is described as the first rack-scale optical inference cluster in a commercial data centre.

The system will use Lumai Iris inference servers to test whether optical computing can cut data centre power use and operate alongside existing AI infrastructure under standard rack, cooling and power conditions.

The deployment puts a UK-developed alternative to conventional silicon-based AI processing into a live test environment as concern grows over artificial intelligence's energy demands. The lab is run by CommonAI, a non-profit group focused on collaborative AI development, and backed by ARIA, the government-supported research funding body.

Lumai's approach performs AI inference using beams of light in three-dimensional space rather than relying solely on transistors in silicon chips. Lumai says this method can reduce data centre energy consumption by 90% and deliver much higher processing speeds, while cutting total cost of ownership to about one-tenth of conventional GPU-based architectures.

Live testing

Researchers will use the cluster to assess the technology under real-world operating conditions. The facility is intended to provide an open environment where academic and industrial teams can test next-generation compute systems on live workloads rather than in controlled laboratory settings.

This matters because the rapid growth of AI inference has intensified concern over data centres' electricity demands. The release cited estimates that meeting the computing ambitions of a single leading AI lab with conventional accelerators alone could produce an infrastructure bill of USD $100 trillion within five years, alongside energy requirements of more than 1,000 GW.

Dr Xianxin Guo, Co-Founder and Chief Executive Officer, Lumai, said: "Power, not demand, now limits how fast AI can be deployed. Running a multi-node system in a commercial data centre is where new compute architectures prove whether they are ready for real deployment. This collaboration at ARIA's Scaling Inference Lab validates our roadmap with a real system operating in a commercial environment."

ARIA selected Lumai after reviewing optical computing systems and concluded that its platform was the most mature of those assessed. The agency was established by an Act of Parliament and is sponsored by the Department for Business, Innovation, Science and Trade.

David Lewis, Chief Science and Technology Officer, ARIA, said: "Lumai's technology is the most mature optical compute platform we evaluated, and has the potential to make AI inference more energy-efficient and less dependent on constrained global hardware supply chains. The Scaling Inference Lab provides the means to test that potential openly and independently at rack scale."

Inference focus

Lumai's hardware is aimed at the "prefill" phase of large-model inference, the stage before the first token is generated and one that depends heavily on matrix computation. Lumai says its system performs millions of matrix operations in a single optical cycle.

Rather than displacing silicon entirely, the servers are designed to work with conventional hardware. In this arrangement, Lumai's system handles prefill while silicon-based processors manage the decode stage, which is more constrained by memory bandwidth.

Lumai says this combination can deliver at least 16 times greater energy efficiency per inference than systems that rely on silicon alone. That claim will now be tested in a rack-scale environment using standard data centre infrastructure.

UK compute push

The project also feeds into a broader debate over the UK's domestic AI computing capacity. With power availability and grid access emerging as practical limits on data centre expansion, supporters argue that more efficient hardware will be needed if the country is to build local compute resources without relying solely on established global chip supply chains.

Gavin Ferris, Chief Executive Officer, CommonAI, said: "As the UK builds out its sovereign AI compute requirements, it has to do so within the power it has. We need to find and then prove new, innovative approaches to AI compute. The Scaling Inference Lab was set up for exactly that - to test innovative technology, such as Lumai's optical compute, on real workloads alongside today's systems."