Hardware/ data-centres · optical-networking · ai-infrastructure · startups

Oriole Networks Says Optical Switches Cut Data Centre Power by 81%

The UK startup, backed by AMD and Aria, wants to replace every electrical switch in the data centre core with nanosecond-scale photonics.

Oriole Networks Says Optical Switches Cut Data Centre Power by 81%

Oriole Networks claims nanosecond-scale optical switches can replace the electrical core of a data centre network and cut switching power by 81%.

The UK company wants to pull out every electrical switch in the network core and drop in photonic alternatives that route data as light rather than converting it back to electrons at each hop. Electrical switches have been the standard for decades: they work, but they consume significant power and generate heat that data centres then pay again to cool. As AI training and inference workloads demand faster, denser interconnects between accelerators, the switching fabric is emerging as a real constraint. Oriole is backed by AMD and Aria, which suggests the pitch has cleared at least some technical due diligence beyond a slide deck.

The 81% figure is the number that matters most and also the one that deserves the most scrutiny. Switching infrastructure is a genuine power draw, but it is one line item in a facility that also runs servers, cooling loops, and power conversion at scale. The headline saving may be smaller as a share of total facility load. That said, optical networking's long-standing problem has been exactly this: every conversion between light and electricity bleeds latency and energy, and a switch that stays entirely in the optical domain sidesteps that. If the nanosecond claims hold under production workloads, it would represent a genuine architectural shift, not a component swap with a better spec sheet.

Plenty of startups have promised to untangle data centre physics before; the interesting test will be what an 81% saving looks like once it leaves a controlled demo and runs inside someone else's rack.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →