Lumai deploys Iris optical servers in ARIA’s £50M Scaling Inference Lab

Lumai’s hybrid optical-digital architecture targets dense matrix operations in AI inference, aiming to execute calculations in parallel via light propagation rather than transistor switching. The company claims potential power savings of up to 90% and total cost of ownership reductions to around one-tenth of conventional GPUs, though these figures are unconfirmed by independent validation. The lab’s structured pilots—each incorporating unproven tech into deployable systems—will compare full-system performance against commercial alternatives, with a new rack planned every six months. Lumai’s CEO, Xianxin Guo, called the ARIA deployment a critical step to prove real-world readiness.