Lambda announced a large-scale deployment of co-packaged optics networking connecting more than 10,000 NVIDIA Blackwell Ultra GPUs at its Quebec campus. The deployment uses NVIDIA Photonics switches and follows Lambda’s November 2025 announcement that it would adopt NVIDIA CPO networking early. Lambda said the deployment will be featured during NVIDIA’s SC26 special address.
What it means
Lambda benchmarked co-packaged optics against conventional pluggable optics at rack, scalable-unit and pod configurations. The company said CPO maintained the same throughput while providing more stable signal integrity and lower latency across those configurations.
Lambda reported that CPO’s latency advantage rose from 3.8 percent at rack scale to 10.1 percent at pod scale. It also reported a 36°C median temperature for CPO optics, compared with 51°C for conventional pluggable optics, and said CPO provides up to 15 times higher reliability.
NVIDIA Photonics switches used in the deployment have 144 800Gb/s ports, liquid cooling and 18 removable external light sources. Lambda’s Quebec campus has 142 MW of secured energy capacity that is nearly 100% renewable, enables free cooling for up to 80% of the year, carries a guaranteed PUE of 1.2 and has no process-water use associated with server cooling.
Key facts
- More than 10,000 NVIDIA Blackwell Ultra GPUs are connected in the deployment.
- Each NVIDIA Photonics switch has 144 ports operating at 800Gb/s.
- The switches include 18 removable external light sources.
- Lambda reported CPO latency advantages of 3.8 percent at rack scale and 10.1 percent at pod scale.
- Lambda reported median optics temperatures of 36°C for CPO and 51°C for conventional pluggable optics.
- Lambda said CPO optics provide up to 15 times higher reliability.
- Lambda’s Quebec campus has 142 MW of secured, nearly 100% renewable energy capacity.
Source: Lambda announcement