Lambda says it is deploying NVIDIA’s co-packaged optics switches at its Quebec campus to connect more than 10,000 Blackwell Ultra GPUs. The claimed gains are lower latency and power use for large AI clusters, where networking can become as consequential as the accelerators themselves.
NVIDIA Watch analysis
What happened
In an announcement published on 16 November, Lambda described what it calls the world’s first large-scale deployment of NVIDIA Photonics switches using co-packaged optics, or CPO. The switches connect AI clusters containing more than 10,000 GPUs at the company’s Quebec campus.
Lambda says the technology reduces latency by up to 10% at pod scale and offers up to five times better power efficiency than conventional pluggable optics. It also claims higher reliability and lower operating temperatures, but the supplied announcement does not give detailed benchmark methods or comparative test conditions.
Why it matters
Moving data between GPUs is part of the cost and performance equation for large AI systems. If CPO delivers the efficiency and latency improvements Lambda describes at this scale, it could make a meaningful difference to operators building and running dense AI clusters. This is a real deployment claim, rather than a lab demonstration, though it is not yet a general verdict on performance across other systems.
Our read
The headline figure is 10,000-plus GPUs, but the interesting test is whether the promised networking gains show up in operating costs and real workloads. “Up to” figures are fine as a starting point; they are not a substitute for the conditions that produced them. We would like to see comparable measurements from more deployments before treating the efficiency claim as a new industry baseline.
What to watch
- Whether Lambda publishes benchmark methods, workload details and measured power use.
- Whether the deployment expands or is followed by other large-scale CPO installations.
- Whether independent customer or operator results support the claimed latency and reliability gains.
Discussion spark: Would up to fivefold better power efficiency be enough to justify adopting co-packaged optics, or should operators wait for comparable results from multiple deployments?
Sources and evidence
- Lambda announces the world’s first large-scale CPO deployment of NVIDIA Photonics switches (16 November 2026, 16:03 UTC)
Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.