Oriole Networks provides PRISM, a pure photonic networking fabric that replaces traditional electrical switches with optical circuit‑switched routers. By moving data with light, PRISM delivers nanosecond‑scale, schedule‑free, contention‑free transfers, supporting up to a million GPU endpoints in a single‑hop, all‑to‑all topology while eliminating power draw and cooling requirements. The solution integrates a full software stack compatible with major AI frameworks, enabling AI and HPC data‑center operators to achieve higher GPU utilization, lower latency, and significantly reduced energy consumption.
Funding
$22M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
4OPPFounders
Product
Problem
Current AI and HPC data centers rely on electrical switching, which consumes large amounts of power, generates heat, and introduces latency and contention that limit GPU utilization and scalability for large language model training and inference.
Solution
Oriole Networks offers PRISM, a pure photonic networking fabric that replaces electrical switches with optical circuit‑switched routers. By moving data with light, PRISM provides schedule‑free, contention‑free transfers with nanosecond switching, dramatically reducing latency and keeping GPUs fully utilized. The architecture scales to up to one million endpoints in a single‑hop, all‑to‑all topology, eliminating congestion and providing dual‑path resilience. Passive optical components eliminate power draw and cooling requirements, lowering overall data‑center energy consumption while supporting both large and small traffic bursts. PRISM integrates via a full software stack—including PCIe drivers, xCCL plugins, and support for PyTorch, TensorFlow, NCCL/RCCL, CUDA, and ROCm—allowing seamless adoption in existing AI workloads.
Target Audience
Primary customers are AI and HPC data‑center operators, cloud providers, and enterprises running large‑scale LLM training, inference, and other compute‑intensive workloads that require high‑performance, energy‑efficient networking.
Features
- Optical circuit‑switched network with schedule‑free, contention‑free data transfers for ultra‑low latency
- Flat one‑hop topology supporting up to 1 million GPU endpoints with all‑to‑all connectivity
- Fully passive photonic routers and switches that consume no power and require no cooling
- Dual‑path, modular design providing resilience without single points of failure
- 800 Gbps pluggable PCIe interface and integrated photonic transceiver (XTR) for wavelength‑aware routing
- Software stack with xCCL plugins, PCIe drivers, and compatibility with PyTorch, TensorFlow, NCCL/RCCL, CUDA, and ROCm
- Bandwidth of 51.2 Tb/s per xPU and 180 ns deterministic latency in PRISM Ultra variant