Oriole Networks is deploying a large-scale AI system with AMD, built on a pure photonic network. The system will form part of the UK's ARIA Scaling Inference Lab.
The project combines Oriole's PRISM networking technology with AMD Instinct GPUs and AMD EPYC CPUs in what the companies describe as the first large-scale deployment of a pure photonic AI network. It is also the first commercial deployment of Oriole's technology.
London-based Oriole says its network replaces electronic switches in the system core with optical circuit switching that routes data as photons. It argues that conventional data centre networks have struggled to keep pace with AI workloads, in which large numbers of chips must exchange data continuously and with low latency.
According to Oriole, the approach cuts core network power consumption by 81% and reduces GPU idle time from about 60% to less than 1%. It says inference throughput increases by an order of magnitude, allowing more tokens per second and more users to be served from the same hardware.
Commercial step
The deployment marks a milestone for the startup, which says it has moved from research and development to production in three years. Its designs are now fixed for a broader industry rollout in 2027 and are intended to work across multiple accelerator platforms rather than being tied to a single chip supplier.
AMD is providing CPU and GPU hardware, along with technical support for the programme. The two companies have worked together for more than a year on large-scale network models linked to AI inference systems.
The work sits within ARIA's Scaling Inference Lab, a testbed backed by £50 million and set up to address constraints in AI workloads. ARIA, the Advanced Research and Innovation Agency, was created by an Act of Parliament and is sponsored by the Department for Science, Innovation and Technology.
Network bottleneck
Data centre operators are under pressure to improve the efficiency of AI infrastructure as demand for training and inference grows. Networking has become a key strain point because accelerators must exchange large volumes of data quickly, while power use and heat output rise as systems scale.
Oriole says its optical design removes the need for electronic switches in the network core, reducing hardware complexity and cooling requirements. It also argues that the architecture could lessen dependence on supply chains tied to existing networking equipment.
For AMD, the project offers a test of how alternative network designs can be integrated with its compute products in AI systems. The chipmaker has been expanding its position in AI hardware, where demand for larger and more efficient clusters has intensified competition across compute, memory and networking.
"AMD is excited to collaborate with Oriole on the ARIA Scaling Inference Lab cluster. Oriole's AI backend networking with nanosecond optical circuit switching represents a fundamentally different way to connect accelerators at scale. 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," said Madhu Rangarajan, Corporate Vice President, Compute and Enterprise AI Business, AMD.
Oriole says the deployment shows that photonic networking can move beyond laboratory work into production systems. The company has positioned its platform as vendor-neutral, arguing that data centre operators will want more flexibility as they mix different compute architectures.
"A year ago, we were proving the physics; today, we're proving the business. 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," said James Regan, Chief Executive Officer, Oriole.
ARIA says the programme is intended to find practical ways to improve the economics and performance of large AI clusters. The agency has focused on inference as a growing challenge as AI services move from model development into wider deployment.
"Meeting the demands for modern AI requires rapidly identifying ways to improve the performance and cost-efficiency of large-scale AI clusters. ARIA is thrilled to collaborate with Oriole and AMD to demonstrate the benefits of this new technology and it's exactly the type of collaboration, between innovative startups and industry leaders, that the Scaling Inference Lab was designed to foster," said Suraj Bramhavar, Program Director, ARIA.