Lambda Deploys World’s First Large-Scale NVIDIA Photonics CPO Network for 10,000+ GPUs

Lambda has switched on the world’s first large-scale co-packaged optics network, connecting more than 10,000 NVIDIA Blackwell Ultra GPUs at its Quebec campus with NVIDIA Photonics switches.

The deployment, announced on Monday, replaces traditional pluggable transceivers with optical engines bonded directly to the switch ASIC. That physical shift — shortening the electrical path from inches to millimeters — cuts network latency by 3.8 percent at rack scale and 10.1 percent at pod scale while matching the throughput of conventional 800 Gb/s optics, Lambda said in its blog.

“Networking has to keep pace with the demands of increasingly large AI clusters,” Dave Ward, president of cloud services at Lambda, told the company’s blog. “Bringing co-packaged optics into a deployment of this scale is an important step toward building the efficient and resilient infrastructure needed to support the next generation of AI.”

Lambda Deploys Nvidia Photonics in Production AI Factory

Gilad Shainer, senior vice president of networking at NVIDIA, framed the economics bluntly: “In an AI factory, the network is a business accelerator. Every interruption stalls the job, and every watt counts.”

Liquid cooling pulls heat from both the switch silicon and the co-packaged optical engines. Lambda reports a median optics temperature of 36 °C versus 51 °C for pluggable modules — a 15 °C drop that translates to more stable signal integrity and, by NVIDIA’s modeling, up to 15 times higher reliability. The design also uses four times fewer lasers, removing a dominant failure domain.

World's first large-scale CPO deployment — Lambda

Each Photonics switch carries 144 ports at 800 Gb/s and 18 removable external light-source modules. Technicians can swap a failed light source in the field without replacing the entire switch — a serviceability improvement over sealed pluggable optics that must be discarded whole when a laser degrades.

Co-packaging eliminates the retimer chips required to drive electrical signals across the longer traces of pluggable designs. NVIDIA claims up to five times better power efficiency per port.

Lambda Boosts AI Factory Resiliency and Efficiency With NVIDIA Photonics
Photo: daily.dev

Lambda says the site runs on 142 MW of nearly 100 percent renewable energy, enables free-air cooling up to 80 percent of the year, and is guaranteed to a power-usage effectiveness of 1.2 with no process water consumed for server cooling.

Lambda first signaled its CPO intent in November 2025, when it announced early adoption of NVIDIA’s co-packaged optics for next-generation AI factories.

Benchmark data released by Lambda shows the latency advantage widening as the fabric scales: 3.8 percent at the rack, 10.1 percent at the pod. Throughput remains identical to pluggable optics across all three tiers, confirming that the signal-integrity gains do not come at a bandwidth penalty.

Lambda’s Dave Ward put the timeline in perspective: “Following its early adoption of NVIDIA CPO networking, first announced in November 2025, Lambda benchmarked the technology against conventional optics, demonstrating lower latency and improved signal stability at scale.”

“The deployment will be featured in NVIDIA’s SC26 special address, offering a closer look at Lambda’s implementation of CPO networking at scale.”

Photo of author

Technology Desk

Technology Desk at Archyde.

Bridgeport Schools Names Dr. Bryan Raach Next Superintendent to Succeed Brent Ripley