Lumai to Test Optical AI Computing at ARIA’s Scaling Inference Lab

Constructor Capital portfolio company Lumai has been selected to deploy its Iris optical inference servers at the UK's Scaling Inference Lab, backed by £50 million from the Advanced Research and Invention Agency (ARIA) and operated by CommonAI CIC.

Lumai will become the lab's first optical inference platform, with its technology evaluated on production-scale AI workloads in a commercial data center. The selection followed a competitive technical evaluation, with ARIA's Chief Science and Technology Officer, David Lewis, describing Lumai as the most mature optical computing platform the agency had assessed.

The deployment marks an important step for Lumai, moving beyond technology demonstrations toward independent, system-level testing under real operating conditions.

Why Optical Computing Matters for AI Infrastructure

As AI inference workloads expand, the economics of compute infrastructure are becoming increasingly important. Power availability, energy efficiency and total cost of ownership affect how much AI capacity data center operators can deploy and at what cost.

Lumai is developing an alternative to conventional silicon-only computing. Its three-dimensional optical architecture uses light to perform the matrix operations central to AI inference, potentially reducing the energy required for compute-intensive workloads.

The company takes a targeted approach. Rather than attempting to replace GPUs entirely, Lumai Iris is designed to accelerate prefill, the computation-heavy stage in which a language model processes its input, while conventional silicon handles decode, the memory-intensive stage of generating the response.

Lumai is targeting up to 90% lower data center power consumption and approximately one-tenth the total cost of ownership of conventional GPU architectures. These remain performance targets that the ARIA deployment will help evaluate under commercial operating conditions.

From Technical Performance to Commercial Viability

For emerging computing architectures, laboratory performance is only part of the challenge. Commercial adoption also depends on whether the technology can integrate into existing infrastructure, operate reliably at scale and deliver measurable economic advantages across the full system.

“This is not a promise of a milestone on a distant roadmap; it’s happening today. Lumai is supplying an optical AI inference cluster to a commercial data center where a third party can operate it. Huge milestone for optical computing!”

Dr. Valentino Jadriško, Senior Associate at Constructor Capital

The Scaling Inference Lab provides an independent environment for testing new AI hardware against commercial systems, using standard data center power, cooling and rack infrastructure.

From an investment perspective, such evaluations help distinguish promising technical capabilities from solutions with a credible path to adoption. Improvements in individual computing operations must ultimately translate into meaningful advantages for infrastructure operators and their customers.

For Lumai, the planned deployment builds on the April 2026 introduction of Iris Nova, its first optical inference server available for evaluation. The next milestone is to demonstrate how the technology performs as part of an integrated, production-scale AI system.

As demand for AI compute continues to grow, the ability to deliver more inference capacity within existing power and cost constraints could become a significant competitive advantage. Independent system-level testing will help establish where optical computing can deliver that value.

Read Lumai's full announcement.