AI & TechArtificial IntelligenceNewswireStartupsTechnology

UK startup cuts data centre power 81% by swapping electrical switches for light

Originally published on: June 8, 2026
▼ Summary

– Traditional data center networks use electrical switches that consume high power, generate significant heat, and create bottlenecks for AI data processing.
– Oriole Networks, a UK startup, proposes replacing every electrical switch in the core network with nanosecond-scale optical switches.

For decades, the internal networks of data centres have relied on electrical switches. These components are notoriously power-intensive, generate significant heat, and have become a primary bottleneck limiting the speed at which AI systems process and exchange information. A UK-based startup, Oriole Networks, claims to have found a solution: replace every electrical switch in the core network with nanosecond-scale optical switches that use light instead of electricity.

This shift could slash data centre power consumption by up to 81%. The company’s approach fundamentally rethinks how data travels between servers, GPUs, and storage systems. Instead of converting optical signals back into electrical ones at every junction,a process that wastes energy and creates heat,Oriole’s photonic network keeps data in the optical domain throughout its journey. The result is not only lower energy use but also reduced latency and higher throughput, which are critical for demanding AI workloads like training large language models and running real-time inference.

The startup has already partnered with AMD and the Aria consortium to test its technology in real-world environments. Early results suggest that the photonic network can match or exceed the performance of traditional electrical networks while using a fraction of the power. As data centre energy demands surge,driven by the rapid expansion of AI,this innovation could become a vital tool for operators seeking to cut costs and meet sustainability targets.

(Source: The Next Web)

Topics

optical networking 95% data center efficiency 90% ai system bottlenecks 88% startup innovation 85% network switches 82% power consumption 80% heat generation 78% nanosecond technology 75% photonic networks 73% data exchange speed 70%