Lumai develops optical compute systems that perform the matrix multiplications used in AI inference with light rather than power-hungry electronic processors. The company, a spin-out from the University of Oxford, is building infrastructure for what it describes as the AI inference era, targeting the energy and cost constraints facing data centres running large language models.
Its product line is the Iris server family, optical compute servers designed to support production-scale large language models in real time. The Iris Nova server, currently being evaluated by hyperscalers, neo-clouds, enterprises and research institutions, encodes vectors and matrix values as light to exploit massive spatial parallelism. Lumai claims substantial improvements in performance, power consumption and total cost of ownership compared with conventional silicon approaches.
Engineering work spans the full stack. Technical domains include optical computing and 3D free-space optics, analog and mixed-signal ASICs, FPGA-based electronics, custom accelerators and optical engines, firmware and drivers, and hardware-specific inference kernels, combining digital and optical processing. The company states a focus on sustainability and energy efficiency in AI computing.
Lumai is based in Oxford, United Kingdom, and serves customers across AI infrastructure, data centres, cloud computing, enterprise AI and research institutions.
