Lumai to deploy world’s first rack-scale optical AI inference cluster in Scaling Inference Lab
In a world first, optical computing company Lumai™ is set to deploy a rack-scale optical AI inference cluster in the Scaling Inference Lab, demonstrating that optical computing can reliably execute large language model (LLM) workloads at scale.
The cluster comprises eight Lumai Iris™ Nova optical inference servers running disaggregated inference alongside NVIDIA GPUs. Iris Nova will handle prefill, the compute-bound stage in which input prompts are processed, while the GPUs handle decode, the memory-bound stage in which output tokens are generated. The deployment aims to run multi-node LLM inference end to end with stable operation under continuous load.
As demand for inference outpaces GPU supply and power, optical computing offers the potential for more power and cost-efficient hardware options and higher throughput.
By performing the matrix multiplications that dominate LLM inference in the optical domain rather than in electronic circuits, Lumai aims to deliver significantly higher performance per watt than state-of-the-art GPUs.
Through partnering with the Scaling Inference Lab, Lumai will gather real-world characterisation data on performance, efficiency and scalability to validate and refine its system simulations, with the lessons gained shaping its next-generation products.
Lumai has already built Iris Nova, which is the world’s first optical computing system capable of running production-scale LLM inference in real time. The Scaling Inference Lab deployment takes this to the next level: proving optical compute at rack scale, on real workloads, in a neutral setting.
“As workloads continue to optimise inference is moving to disaggregated architectures, where prefill and decode each run on specialised hardware. The Scaling Inference Lab will demonstrate optical compute-based prefill alongside decode processors, at rack scale and on real workloads. The data from the cluster will characterise the performance and scaling of optical compute across multiple nodes, and demonstrate successful optical computing in a third-party data center.”
Phillip Burr — Head of Product, Lumai
“By demonstrating optical computing in this real-world environment, the cluster will uncover critical insights for next-generation AI infrastructure, accelerating commercial adoption and shaping the future of high-performance, energy-efficient AI inference.”
Dr. Rosemary Francis — CTO, CommonAI Compute
The deployment will generate calibrated benchmark data to test Lumai’s projections of up to a 90% reduction in power consumption compared with silicon GPUs, a 50x increase in throughput for high-batch inference workloads, and a significantly lower total cost of ownership.
The mission behind the Scaling Inference Lab
The Scaling Inference Lab provides a testbed for new AI technologies, prioritising rapid iteration, open collaboration, and long-term sustainability.
The Scaling Inference Lab is a £50m programme funded by ARIA (The Advanced Research and Invention Agency), backed by a further £25m from BIST (Department for Business, Innovation, Science and Trade). The Lab aims to provide shared software and hardware infrastructure that lowers the barrier for innovative startups to plug into larger rack-scale systems and ensures they are tested against the latest emerging workloads.
Learn more about the Scaling Inference Lab and the hypotheses being tested on the website: scalinginference.org
About CommonAI C.I.C.
CommonAI C.I.C. is a non-profit membership organisation, founded on a belief in collaborative engineering for the safe and responsible development of foundational AI technologies. We are focused on building shared infrastructure that organisations can use to run and improve AI systems in real conditions with sovereignty, certainty and efficiency.
About Lumai
Lumai is the optical compute company advancing energy-efficient AI inference with the unique power of light. Its 3D optical computing architecture accelerates compute-intensive inference workloads beyond the energy constraints of conventional compute, helping data centers increase AI capacity within power budgets. Iris Nova, the first server in the family, is available for evaluation today. Spun out of world-leading optics research at the University of Oxford in 2021, Lumai’s mission is to unlock sustainable intelligence at global scale. Lumai's work has been recognized with the Falling Walls Award for Science Breakthrough of the Year 2025 and Best Overall Technology at the OCP Future Technologies Symposium, and the company is part of Intel Ignite's first London cohort. Visit Lumai at www.lumai.ai.
For media enquiries, please contact: aniewiadomska@commonai.org