SourceThe Next Web (TNW)•Stalecollected in 2h
UK startup claims 81% power cut with optical switching
💡A potential 81% power reduction in data centers could redefine AI infrastructure economics.
⚡ 30-Second TL;DR
What Changed
Oriole Networks replaces electrical switches with optical ones
Why It Matters
If scalable, this could significantly lower the operational costs and environmental footprint of large-scale AI training clusters.
What To Do Next
Keep an eye on optical networking hardware benchmarks; this could be the next major efficiency leap for your GPU cluster architecture.
Who should care:Developers & AI Engineers
Key Points
- •Oriole Networks replaces electrical switches with optical ones
- •Claims an 81% reduction in data center network power consumption
- •Addresses heat and bandwidth bottlenecks in AI systems
🧠 Deep Insight
Background and context from public sources — not the original article. 17 sources cited.
🔑 Enhanced Key Takeaways
- •Oriole Networks' PRISM platform is being deployed in collaboration with AMD hardware, specifically Instinct GPUs and EPYC CPUs, within the UK's £50 million ARIA Scaling Inference Lab, marking its first commercial use.
- •The optical switching technology is projected to reduce GPU idle time from approximately 60% in conventional systems to less than 1%, which Oriole claims will result in an order-of-magnitude increase in AI inference throughput.
- •Founded in 2023 as a spinout from University College London (UCL), Oriole Networks builds upon more than two decades of optical networking research conducted by co-founder Professor George Zervas and his team.
- •Oriole Networks has secured a total of $34.6 million across two funding rounds, including a $12.6 million seed round and a $22 million Series A round led by Plural.
- •The PRISM solution is designed to support systems with up to one million endpoints with a 1-hop diameter, offering true all-to-all connectivity and handling both deterministic collective communications and non-deterministic dynamic traffic.
🛠️ Technical Deep Dive
- Oriole's PRISM platform completely removes electronic packet switches from the network core, replacing them with nanosecond-speed optical circuit switching.
- Data is routed as photons directly from chip to chip, eliminating optical-to-electrical conversions.
- The full-stack solution comprises a high-performance Network Interface Card (providing 800 Gbps bandwidth), an Integrated Photonic Switch and Transceiver (XTR) for on-the-fly wavelength and route selection, and a passive, modular Photonic Router.
- It utilizes nanosecond-level switching and network control across time, wavelength, and space (TDM, WDM, and SDM) to enable rapid circuit configuration, efficiently managing both large and small data transfers.
- The network core features entirely passive interconnects and switches, which means it does not consume power or require cooling.
- PRISM offers true all-to-all connectivity with a 1-hop diameter, capable of scaling to systems with up to one million endpoints.
- The architecture integrates the physical network, AI communication models, and scheduling logic, employing purpose-built algorithms for collective communication strategies optimized for optical circuit-switched networks.
- Its software stack includes xCCL plugins and PCIe drivers designed for seamless integration with existing infrastructure.
- An announced variant, PRISM Ultra, claims a throughput of 50 Exabit per second and 180ns xPU to xPU latency, achieving extensive connectivity with less than 25% of the total system power.
🔮 Future ImplicationsAI analysis grounded in cited sources
Oriole Networks' technology will significantly accelerate the adoption of optical networking in AI data centers.
Its demonstrated power efficiency and performance gains, coupled with a commercial deployment with AMD in a UK national lab, provide strong validation for broader industry rollout.
The widespread adoption of photonic networking could lead to a substantial reduction in the environmental footprint of AI infrastructure.
By drastically cutting power consumption and cooling demands, Oriole's solution directly addresses the increasing energy and water usage of large-scale AI systems.
Oriole's vendor-agnostic PRISM platform will foster greater competition and innovation at the GPU layer.
By not being tied to any single chip vendor and running across various accelerator platforms, it offers customers more flexibility and reduces reliance on proprietary networking stacks.
⏳ Timeline
2023
Oriole Networks founded as a spinout from University College London.
2024-03
Raised a Seed funding round of $12.6 million (or £10 million).
2024-10
Raised a $22 million Series A funding round led by Plural, bringing total funding to $34.6 million.
2025-05
Unveiled PRISM (Photonic Routing Infrastructure for Scalable Models), described as the world's first full-photonic network for AI.
2026-03
Announced PRISM Ultra, a one-hop photonic network fabric with 50 Exabit per second throughput.
2026-06
Deployed the world's first large-scale AI system powered by a pure photonic network in collaboration with AMD, as part of the UK's ARIA Scaling Inference Lab, marking its first commercial use.
📎 Sources (17)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: The Next Web (TNW) ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.