Oriole says it will deploy first pure photonic AI network with AMD in UK lab

Oriole Networks said it is moving its photonic AI networking system into a UK government-backed testbed with AMD support, aiming to cut latency and power use in data centers. The collaboration is tied to ARIA’s Scaling Inference Lab and is positioned as the first commercial deployment of Oriole’s technology.

Why it matters: - Oriole’s system targets one of AI’s biggest infrastructure limits: the network bottleneck between chips. - The company says photonic switching can cut core power consumption, reduce GPU idle time, and let the same hardware serve more users and generate more tokens per second. - The deployment matters for UK AI infrastructure because ARIA’s Scaling Inference Lab is designed to test technologies that improve performance and efficiency at scale.

What happened: - Oriole Networks announced continued progress in its collaboration with AMD and the UK’s Advanced Research & Invention Agency, or ARIA. - The work combines Oriole’s photonic networking system with AMD Instinct GPUs and AMD EPYC CPUs. - The collaboration has been underway for more than a year. - Oriole says it is set to deploy the world’s first pure photonic AI network at scale. - The company said the deployment is intended to supercharge AI performance at the system level by delivering the lowest possible latency. - Oriole said this marks the first commercial deployment of its technology. - The company said its xPU-agnostic designs are locked and set for wider industry rollout in 2027.

The details: - Oriole’s PRISM platform replaces electronic switches in the network core with nanosecond-scale optical circuit switching. - The company says PRISM routes data as photons rather than electrical signals. - Oriole says the system cuts core power consumption by 81%. - Oriole says GPU idle time falls from 60% today to less than 1%. - The company says the design can reduce cooling needs and water use. - Oriole says the approach can also reduce dependence on the supply chain that supports traditional networking hardware. - AMD is providing CPU and GPU hardware plus technical collaboration to develop and run large-scale network models for frontier-scale AI systems. - Madhu Rangarajan, corporate vice president for AMD’s Compute and Enterprise AI business, said AMD is helping validate how photonic fabrics can work alongside AMD compute to deliver low-latency, high-bandwidth connectivity for AI inference workloads. - The Scaling Inference Lab is backed by £50 million, or $68 million. - ARIA was created by an Act of Parliament and is sponsored by the UK Department for Science, Innovation and Technology. - UK Business and Trade Secretary Peter Kyle said the system could help attract investment, support skilled job growth, and improve the ability of UK scale-ups to compete globally.

Between the lines: - Oriole is framing photonic networking as a shift from research to production, not just a lab experiment. - The AMD partnership gives the deployment more credibility because it links a startup’s networking architecture with mainstream compute hardware. - The focus on vendor-agnostic design suggests Oriole wants broader adoption across accelerator platforms, not a single-chip ecosystem. - The claims about power, latency, and throughput point to a broader race to make AI clusters cheaper and more efficient as model sizes and inference demand keep rising.

What’s next: - Oriole said its xPU-agnostic designs are expected to roll out more widely across the industry in 2027. - The ARIA lab will continue serving as a testbed for measuring whether photonic networking can improve AI cluster performance at frontier scale. - Oriole and AMD will keep developing and validating the network models used in the collaboration. - The company’s next proof point will be whether the deployment delivers the performance gains and efficiency savings it is promising at larger scale.

The bottom line: - Oriole is trying to turn photonic networking from a lab concept into core AI infrastructure, with AMD and ARIA providing the platform for a first commercial-scale test.