Lambda announced on November 16, 2026, that it has deployed a network using NVIDIA Photonics switches to connect more than 10,000 NVIDIA Blackwell Ultra GPUs at its Quebec campus. The company calls it the world’s first large-scale deployment of co-packaged optics, or CPO. CPO places optical components alongside networking chips, using light to carry data between systems. The announcement concerns how the GPUs are connected, not a new GPU model or AI service.
Lambda says it compared the CPO network with conventional optics and found lower latency and more stable signals at scale. Latency is the delay before data arrives; lower latency can matter when many processors need to exchange information during a computing job. The company did not provide figures for the size of the latency reduction, the stability measurements, or the benchmark setup in the supplied announcement. That means readers cannot use the claim to estimate how much faster a particular workload would run.
What the deployment means
Lambda says its adoption of NVIDIA CPO networking was first announced in November 2025, making this announcement a report of a large-scale deployment following that earlier step. The new details identify the location, the switch technology and the number of GPUs connected. They do not say when customers can use the system, which services will run on it, or whether access will be offered outside Lambda’s cloud. The campus deployment establishes a change in Lambda’s infrastructure, but the announcement does not describe a new product that individual customers can buy or enable themselves.
For organizations running large AI workloads, a network connecting more than 10,000 GPUs is relevant because it links a sizable pool of computing hardware. Lambda’s reported latency and signal-stability results suggest the company sees CPO as useful at that scale. But without published measurements or information about customer access, it is not possible to tell what performance benefit a user would experience, or whether the deployment changes the cost or availability of Lambda’s services.
Why it matters
For companies using Lambda’s computing infrastructure, the deployment connects a large pool of GPUs and may improve how quickly they exchange data, according to Lambda. The announcement gives no benchmark figures or customer-access details, so the practical performance benefit and who can use the system remain unclear.