Oriole Networks Deploys Photonic AI Network in First Commercial Rollout with AMD

Why It Matters

The photonic network dramatically cuts energy and cooling costs while boosting AI inference efficiency, giving data‑center operators a new path to scale workloads sustainably. Its vendor‑agnostic design also lowers reliance on proprietary networking, accelerating broader adoption across the AI hardware ecosystem.

Key Takeaways

• Oriole's PRISM uses photons, replacing electronic switches in data centers.
• Deployment cuts core power by 81% and GPU idle time below 1%.
• AMD supplies Instinct GPUs and EPYC CPUs, supporting photonic AI network.
• Scaling Inference Lab funded with $66.6 million to accelerate AI infrastructure.
• Oriole raised $35 million, targeting broader rollout by 2027.

Pulse Analysis

The explosion of generative AI has exposed a fundamental bottleneck in traditional data‑center architecture: electrical switching fabrics that consume massive power and generate heat. Photonic networking, which transmits data as light rather than electrons, promises nanosecond‑scale latency and near‑zero energy loss, but scaling the technology from laboratory prototypes to production has remained elusive. Oriole Networks, a spin‑out from University College London, has spent the past three years refining its PRISM platform to address these constraints, positioning itself at the forefront of a nascent market that could redefine how AI workloads are interconnected.

In collaboration with AMD, Oriole has now integrated PRISM with Instinct GPUs and EPYC CPUs in the UK’s Scaling Inference Lab, a $66.6 million initiative backed by the Advanced Research and Invention Agency. The deployment replaces conventional electronic switches with optical circuit switches, slashing core power draw by 81% and reducing GPU idle time from roughly 60% to under 1%. By eliminating the heat‑intensive switching layer, the system also cuts cooling requirements and water usage, delivering a leaner, faster substrate for inference tasks that must move terabytes of data trillions of times per second.

The commercial rollout signals a turning point for the broader AI infrastructure ecosystem. With a vendor‑agnostic photonic fabric, customers can avoid lock‑in to a single chip maker, fostering a more competitive hardware landscape. Oriole’s $35 million funding round, led by investors such as Plural UK and XTX Ventures, underscores confidence in the technology’s scalability, with a full industry release slated for 2027. As hyperscale operators chase lower TCO and greener operations, photonic AI networks could become a standard component, accelerating the next wave of AI innovation while delivering measurable sustainability gains.