U.K. photonic networking startup Oriole Networks Ltd. today announced that it will deploy what it describes as the world’s first large-scale artificial intelligence system running on a pure photonic network, marking the first commercial use of its technology.

The deployment, built in collaboration with Advanced Micro Devices Inc., forms part of the U.K. Advanced Research and Invention Agency’s Scaling Inference Lab, a testbed backed by £50 million ($66.6 million) set up to tackle bottlenecks in large-scale AI workloads. The system pairs Oriole’s networking hardware with AMD Instinct graphics processing units and AMD EPYC central processing units.

Oriole is contributing PRISM, a networking platform that routes data as photons rather than electrical signals. The system replaces the electronic switches at the core of a data center network with nanosecond-scale optical circuit switching. AMD is supplying the compute hardware and technical support to build and run network models at frontier scale.

Conventional data center networks rely on electrical switches that draw heavy power and generate substantial heat. That has become a growing constraint as AI workloads push thousands of chips to exchange data trillions of times per second. Oriole says removing electronic switches cuts core power consumption by 81% and drops graphics processing unit idle time from about 60% today to less than 1%. It says the approach also lowers cooling demands and cuts water use.

PRISM is not tied to any single chip vendor and runs across accelerator platforms, Oriole said, a selling point for customers that want to avoid proprietary networking stacks. The company said its designs are now locked for wider industry rollout in 2027.

Oriole has been working with AMD for more than a year. Founded in 2023 as a University College London spinout, the company said its technology went from research to production in three years.

“A year ago, we were proving the physics, today, we’re proving the business,” said Chief Executive James Regan. “Our collaboration with AMD has moved from concept to deployment to a system an order of magnitude larger and the data proves this is already driving performance increases at pace.”

Madhu Rangarajan, corporate vice president of compute and enterprise AI at AMD, said in the announcement that Oriole’s nanosecond optical circuit switching represents a fundamentally different way to connect accelerators at scale.  He added that AMD is helping validate how photonic fabrics can deliver the low-latency, high-bandwidth connectivity that AI inference workloads demand.

ARIA, created by an act of Parliament and sponsored by the Department for Science, Innovation and Technology, funds early-stage research aimed at long-term economic gains. Program director Suraj Bramhavar said the Oriole work is the type of partnership between startups and established players that the Scaling Inference Lab was set up to foster.

Oriole has raised about $35 million in funding to date. Investors in the company include Plural UK Management Ltd., UCL Technology Fund, Clean Growth Fund, XTX Ventures and Dorilton Ventures.

Image: Oriole