AiGpu

Infrastructure·

Lambda Deploys NVIDIA Photonics Co‑Packaged Optics to Boost AI Factory Resilience

Lambda integrates NVIDIA Photonics co‑packaged optics switches across more than 10,000 GPUs, cutting latency, power draw, and failure rates for large‑scale AI training.

NVIDIA Photonics co‑packaged optics switch deployed in Lambda data center

Lambda, a specialist GPU cloud provider, has rolled out NVIDIA Photonics co‑packaged optics (CPO) switches to interconnect its NVIDIA GB300 NVL72 racks, linking over 10,000 GPUs in a single AI factory.

In benchmark runs the new switches shaved roughly 4 % off rack‑level latency and up to 10 % at the pod level while preserving the same throughput as traditional pluggable optics.

The CPO design delivers up to five times better power efficiency and fifteen times higher reliability than conventional transceivers, using four times fewer lasers and eliminating retimer chips from the signal path.

Liquid‑cooled switch packages keep median optics temperature at 36 °C versus 51 °C for pluggable modules, and the 18 field‑replaceable external light sources let technicians swap a failed laser without removing the entire switch.

Why it matters for GPU / AI infrastructure

Network interruptions in a 10,000‑GPU cluster can stall an entire training run, wasting both compute cycles and energy. By lowering latency, cutting power per bit, and dramatically improving mean time between interruptions, co‑packaged optics directly translate into higher GPU utilization and lower total cost of ownership for AI factories.

  • aigpu
  • ai gpu
  • ai gpu cloud
  • aigpu dubai
  • nvidia photonics
  • co-packaged optics
  • lambda
  • ai factory

By AiGpu Editorial · Editorial rewrite based on public reporting (NVIDIA Blog)

← All articles