Oriole to deploy photonic AI network
Oriole Networks will deploy what it describes as the world’s first pure photonic AI network at scale, combining its optical networking technology with AMD processors to improve AI system performance and efficiency.
Oriole Networks has announced plans to deploy a large-scale photonic AI network as part of its ongoing collaboration with AMD and the U.K.’s Advanced Research & Innovation Agency (ARIA) Scaling Inference Lab.
The project will combine Oriole’s PRISM (Photonic Routing Infrastructure for Scalable Models) networking platform with AMD Instinct GPUs and AMD EPYC CPUs to demonstrate next-generation network fabrics designed to address performance, latency, and energy challenges in AI infrastructure.
According to Oriole, PRISM replaces electronic switches in the network core with nanosecond-scale optical circuit switching, enabling direct photonic communication between computing resources.
The company said the approach is designed to reduce latency and improve utilization across large-scale AI systems.
The deployment marks the first commercial implementation of Oriole’s technology.
The company stated that it has progressed from research and development to production within three years and plans a broader industry rollout of its accelerator-agnostic networking architecture in 2027.
Oriole claims that replacing electronic switching infrastructure with optical circuits can reduce core network power consumption by up to 81%.
The company also estimates that GPU idle time can be reduced from approximately 60% to less than 1% by enabling more efficient communication between processing elements.
In addition to performance gains, the photonic networking architecture is intended to reduce reliance on complex electronic networking hardware and lower cooling requirements, potentially decreasing overall energy and water consumption in AI data centres.
The collaboration highlights growing industry interest in photonic networking technologies as AI workloads continue to scale.
By leveraging optical communication at the system level, developers aim to overcome limitations associated with conventional electronic interconnects and support the increasing computational demands of next-generation AI models.
