Lumai™ states that using light for AI inference can reduce data center power consumption by 90% and increase processing speed by 50 times.
Oxford, UK, November 3, 2026 /PRNewswire/ — The UK’s public R&D funding agency, the Advanced Research and Invention Agency (ARIA), will deploy Lumai’s 3D optical reasoning technology for the first time. Lumai is an optical computing company building the next generation of AI infrastructure.
As the issue of data center energy consumption grows in importance on political and industrial agendas, ARIA will integrate the Lumai Iris™ inference server into its Scaling Inference Lab as the lab’s first optical inference platform. Operated by the nonprofit organization CommonAI CIC, which focuses on collaborative AI development, this deployment will enable researchers to evaluate the value of optical computing in commercial data center environments and its potential to reduce data center power consumption by 90%. The Lumai Iris server may also reduce total cost of ownership to approximately one-tenth that of traditional GPU architectures.
Lumai’s Iris technology can run production-grade large language models (LLMs) by performing AI computations within beams of light in three-dimensional space, rather than relying solely on transistors in silicon chips. Since numbers are encoded as light intensity and passive optical components like lenses are used for mathematical operations, the computation itself consumes almost no energy, and energy consumption barely increases as performance improves.
“Electricity, not demand, is now limiting the speed of AI deployment,” said Dr. Xianxin Guo, co-founder and CEO of Lumai. “Running multi-node systems in commercial data centers is precisely where new computing architectures prove whether they are ready for real-world deployment. This collaboration with ARIA Scaling Inference Lab validates our roadmap on a real system in a commercial environment.”
ARIA is an independent funding agency supported by the UK government, dedicated to advancing artificial intelligence and other transformative technologies. Established by an Act of Parliament and sponsored by the UK Department for Business, Innovation, Science and Trade, ARIA supports scientists and engineers in pursuing ambitious ideas with the potential to deliver major scientific and technological breakthroughs—particularly those that benefit society at large. Its Scaling Inference Lab aims to stress-test next-generation computing architectures in real-world commercial environments using standard power, cooling, and rack integration.
David Lewis, Chief Scientific and Technology Officer at ARIA, said: “Among the platforms we have evaluated, Lumai’s technology is the most mature optical computing platform with the potential to make AI inference more energy-efficient and reduce dependence on constrained global hardware supply chains. The Scaling Inference Lab provides a means to openly and independently test this potential at rack scale.”
The Path to More Efficient Inference
As demand for AI inference accelerates, building data center efficiency will become critical. Traditional AI inference operates electronically on silicon and is highly power-intensive at scale. If we rely solely on these traditional accelerators to meet the computational demands of even a leading AI lab, infrastructure spending is projected to reach $10 trillion over the next five years. Additionally, the extra energy demand would exceed 1,000 gigawatts. Without changing the physical approach to AI computation, these energy and cost requirements would make achieving this goal impossible.
Sovereign AI requires more intelligent computing
For the UK, this challenge also concerns sovereignty. Building the domestic AI computing capacity needed by the UK will depend on extracting more computing power within limited electricity and grid capacity, and bringing UK-developed technologies into real-world data centers.
Gavin Ferris, CEO of CommonAI, said: “As the UK builds its sovereign AI computing capabilities, it must do so within existing power constraints. We need to identify and validate new, innovative approaches to AI computing. The Scaling Inference Lab was created for exactly this purpose—to parallel-test optical computing innovations like Lumai against today’s systems under real-world workloads.”
Handle compute-intensive prefill workloads
Large model inference consists of two stages: the prefill stage is compute-bound, while the decode stage is memory-bandwidth-bound. Although dedicated inference hardware is increasingly focusing on the decode stage, Lumai takes a different approach—it targets the prefill stage, which is computationally intensive and involves heavy matrix operations before generating the first token. Lumai Iris can perform millions of matrix operations within a single optical cycle.
The Iris server is not a replacement for silicon chips, but rather excels when working alongside them during decode processing—achieving at least a 16-fold improvement in energy efficiency per inference compared to silicon-only systems, thereby addressing the power and cost constraints currently limiting AI deployment.
ARIA's deployment builds upon Iris Nova, launched in April 2026. Iris Nova is Lumai's first product available for evaluation and the first optical AI accelerator designed specifically for data center-scale inference.
About Lumai
Lumai is an optical computing company leveraging the unique properties of light to advance energy-efficient AI inference. Its 3D optical computing architecture accelerates compute-intensive inference workloads, overcoming the power limitations of traditional computing and enabling data centers to increase AI capacity within their power budgets. Iris Nova, the first server in this series, is currently available for evaluation. Lumai was spun out from world-leading optical research at the University of Oxford in 2021, with a mission to unlock sustainable intelligence at global scale. Lumai’s work has been recognized with the 2025 Falling Walls Award for Scientific Breakthrough and the OCP Future Technologies Symposium Best Overall Technology Award, and the company is a participant in the inaugural Intel Ignite London program. Visit Lumai at www.lumai.ai.
About ARIA
The Advanced Research and Invention Agency (ARIA) is a UK research funding body dedicated to unlocking scientific and technological breakthroughs that benefit everyone. By empowering scientists to explore the boundaries of possibility, ARIA drives ambitious projects that catalyze progress and create new communities and industries. The agency brings together scientists and engineers with diverse expertise, breaking down traditional barriers to open new pathways for technological advancement. https://aria.org.uk
About Scaling Inference Lab
The Scaling Inference Lab, funded with £50 million by ARIA and implemented by CommonAI CIC, aims to address a key bottleneck faced by startups and research teams developing innovative AI hardware: the lack of an open platform to showcase their technologies in today’s AI hardware market, which is dominated by large closed systems.