AMD is working with British photonics firm Oriole Networks on what the pair claim is the world’s first pure photonic AI network.

The effort, which has been in development for more than a year, aims to pair Oriole’s photonic networking system with Instinct graphic processing units (GPUs) and EPYC central processing units (CPUs) to showcase how next-generation network fabrics could use light to transmit data rather than traditional electrical signals.

Oriole’s Photonic Routing Infrastructure for Scalable Models (PRISM) technology replaces electronic switches with nanosecond-switched optical circuits. While microelectromechanical systems (MEMS) require milliseconds to configure routing, Oriole placed its bets on a technology that changes routing pathways at the scale of a billionth of a second.

Microsoft researchers previously explored nanosecond optical switching for data centers through its Sirius project, contending that MEMS-based optical architectures have risen in popularity, but are slower than the packet-granularity reconfiguration offered by electrical packet switches for smaller network flows – a key consideration with the emergence of AI-induced infrequent, burst-based workloads.

Oriole’s PRISM is a full-stack offering made up of 800 Gb/s (800G) network interface cards (NICs), modular photonic routers, integrated photonic switches, and transceivers, all of which are tied together via its software stack that substitutes default transport layers with optimized plugins and drivers. An Ultra version of PRISM was revealed back in March, which transforms the stack into a flat one-hop fabric that “scales to a million XPUs.”

The University College London spinout argues its core nanosecond optical switching platform can cut network power consumption by 81%, while GPU idle time drops from 60% today to less than 1% as photons are able to travel directly from chip to chip. The firm also makes the case that PRISM can reduce dependency on complex networking hardware supply chains and minimize the need for cooling.

“A year ago, we were proving the physics; today, we’re proving the business,” Oriole CEO James Regan noted. “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. This is what it looks like when photonic networking stops being a research curiosity and starts being the foundation of how serious AI infrastructure gets built.”

Oriole is also pitching its PRISM photonic networking technology as suitable for any accelerator platform, which makes ample sense given its support from heterogeneous computing cheerleader AMD that is vying for frontier-scale system-wide performance without the need for proprietary stacks. Oriole’s XPU-agnostic designs are expected for wider rollout in 2027.

“Oriole’s AI backend networking with nanosecond optical circuit switching represents a fundamentally different way to connect accelerators at scale,” Madhu Rangarajan, corporate VP for AMD’s compute and enterprise AI business, added. “We are helping to validate how photonic fabrics can work alongside AMD compute to deliver the low-latency, high-bandwidth connectivity that AI Inference workloads demand.”

Oriole was founded in 2023 with its technology based on research co-founder George Zervas has iterated on for the past two decades. Crunchbase figures state the firm has raised around $35 million, securing capital from backers including XTX Ventures and UCL’s Technology Fund.

AMD’s involvement came as the chip giant revealed it’s investing up to $2.6 billion over the next five years in the U.K., supporting local research efforts and “accelerat[ing]” the nation’s AI ecosystem.

“The U.K. has the talent, research excellence, and ambition to help lead the next era of AI,” AMD CEO Lisa Su noted. “AMD is proud to deepen our commitment to the U.K. and work with partners across government, academia, and industry to expand access to the compute infrastructure needed to advance sovereign AI, accelerate discovery, and drive long-term economic growth.”