Lambda, a specialist in artificial intelligence infrastructure, has announced the operational launch of the world’s first large-scale deployment of co-packaged optics (CPO) networking, connecting over 10,000 NVIDIA Blackwell Ultra GPUs at its Quebec-based campus. The milestone, confirmed by Lambda on November 16, 2026, marks a leap in data center networking, enabling more efficient, reliable, and scalable AI training and inference at industrial scale.

This deployment, built on NVIDIA’s photonics networking switches with CPO technology, is a response to surging computational demands in large neural networks. In internal testing, Lambda reports that the new switches, each with 144 ports at 800 Gb/s, delivered a reduction in network latency by up to 10% at the pod scale compared to conventional pluggable optics, while also improving signal integrity and cooling efficiency. Median operating temperatures for the optics are down to 36°C, significantly lower than the 51°C seen with standard solutions—a critical performance margin in energy-intensive AI deployments.

Co-packaged optics, which integrate optical connections directly on the networking chips, represent a substantial architectural evolution for AI data centers. The design offers not only enhanced throughput but a step-change in reliability, which Lambda quantifies as up to 15 times greater than traditional optics modules. In a further practical innovation, each NVIDIA Photonics switch uses 18 removable external light sources, simplifying maintenance and reducing downtime for large cloud installations.

The new infrastructure is cooled by advanced liquid cooling systems, an approach regarded as increasingly necessary as centers for AI computation grow in density and complexity. Lambda’s operational deployment builds on the foundation laid by NVIDIA’s introduction of its Spectrum-X Photonics line in 2025, which first demonstrated the feasibility of energy-efficient, co-packaged optics at scale.

While Lambda’s early performance results are promising, company representatives have acknowledged that longer-term durability and broader scalability across additional campuses will remain key points of observation. Nonetheless, industry observers see the move as a significant harbinger of a shift for AI infrastructure: as models become larger and demands intensify, advanced networking architectures such as these may transition from cutting-edge trial to sector-wide necessity.

The Quebec site’s deployment is now live and supporting Lambda’s roster of AI applications. No conflicting reports have emerged to dispute Lambda’s claims, and the industry is watching closely as other providers consider similar advances in photonics-enabled networking to fuel the next generation of AI capabilities.