ARIA will test Lumai Iris optical AI servers at rack scale
ARIA’s Scaling Inference Lab, operated by CommonAI, selected Lumai Iris as its first optical inference platform for evaluation on real workloads in a commercial data center.
AT A GLANCE
Key points
- Lumai Iris will be the lab’s first optical inference platform.
- Testing will use commercial data-center racks, power and cooling.
- The 90% power-saving figure is a provider target, not an independent result.
What is changing?
Lumai announced on November 3 that its Iris optical inference servers will join the UK Advanced Research and Invention Agency’s Scaling Inference Lab. Operated by CommonAI CIC, the lab will test Iris at rack scale in a commercial data-center environment as its first optical inference platform.
Iris is designed to run matrix operations for the compute-heavy prefill stage of large-language-model inference using three-dimensional beams of light, while working alongside silicon accelerators for memory-heavy decoding. The evaluation will use real workloads with standard power, cooling and rack integration.
Lumai says the technology could reduce data-center power by up to 90% and bring total cost of ownership to about one tenth of conventional GPU architectures. These figures are company targets, not independent laboratory results.
Explore the original sources
Source date: 3 November 2026.
Published by AI Radar: 4 Nov 2026, 02:16 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…