Lambda says it has put NVIDIA Photonics co-packaged optics switches into service to connect NVIDIA GB300 NVL72 racks with more than 10,000 GPUs at its Quebec campus. NVIDIA’s 16 November blog post and Lambda’s same-day announcement describe the installation as a large-scale deployment of co-packaged optics, or CPO, in which optical engines sit alongside the switch chip rather than in conventional pluggable transceivers at the front panel. Lambda called it “the world’s first large-scale deployment” of the technology. That “first” is the company’s claim. Both announcements also describe the technical change itself: CPO shortens the electrical path from inches to millimetres, which, they say, reduces signal loss and removes the need for retimer chips.
In Lambda’s tests, the NVIDIA blog said, network latency “dropped by about 4% at rack scale and by up to about 10% at pod scale,” with the same throughput as conventional pluggable optics across rack, scalable-unit and pod configurations. Lambda’s own announcement gave the same tests in more precise terms, saying the latency gap “growing from 3.8 percent at rack to 10.1 percent at pod.” Those figures come from Lambda’s tests, not from an independent benchmark. Lambda also reported a median optics temperature of 36°C, compared with 51°C for conventional pluggable optics, and said its tests found more stable signal integrity.
Dave Ward, president of cloud services at Lambda, said: “Networking has to keep pace with the demands of increasingly large AI clusters. Bringing co-packaged optics into a deployment of this scale is an important step toward building the infrastructure needed to support the next generation of AI.”
What NVIDIA claims for the switches, and what Lambda says about the site
Each NVIDIA Photonics switch, the companies said, has 144 ports running at 800Gb/s, is liquid-cooled, and uses 18 removable external light sources that supply the optical engines and can be replaced in the field. NVIDIA said that by replacing pluggable transceivers with co-packaged optical engines, the design delivers “up to 5x better power efficiency and up to 15x higher reliability” than conventional pluggable optics, and uses “4x fewer lasers.” Those multipliers are NVIDIA’s claims. Lambda repeated the “up to 15x higher reliability” figure in its own release, tying it to cooler, more stable optics.
Lambda said the Quebec campus has 142 MW of secured, nearly 100% renewable energy capacity, can use free cooling for up to 80% of the year, and is guaranteed to a power-usage effectiveness of 1.2, with no process water consumption associated with server cooling. Those are the company’s figures for the site. It also said the deployment will be featured in NVIDIA’s SC26 special address, and that it first announced early adoption of NVIDIA CPO networking in November 2025.
The installation sits beside other recent TSN accounts of large AI-factory and optical-networking plans, including Lambda’s Alabama campus, pegged at a $33.1 billion investment, Salience Labs’ funding to scale optical switches for AI data centres, and PsiQuantum’s work on an optical circuit switch. Those are neighbouring infrastructure stories, not this Quebec deployment.
Lambda’s test figures are not a general networking law
The latency numbers that matter in this announcement are Lambda’s. The 3.8% rack-scale and 10.1% pod-scale figures, and the rounded “about 4%” and “up to about 10%” versions in NVIDIA’s blog, are results the company reports from its own comparison with conventional pluggable optics. So are the 36°C and 51°C median temperatures. They describe this deployment’s tests, not a guaranteed improvement for every CPO cluster.
NVIDIA’s “up to 5x” power-efficiency, “up to 15x” reliability and “4x fewer lasers” claims should be kept separate. They are vendor performance statements about the Photonics switches, not measurements independently confirmed in Lambda’s published figures, even though Lambda repeats the reliability multiplier. The “world’s first large-scale deployment” label is likewise Lambda’s description, not a third-party ranking.
The “more than 10,000 GPUs” count, the 144 ports and the 18 removable light sources are the companies’ specifications for this installation. Until another operator, or a later Lambda disclosure, publishes a comparable test, the documented result is a vendor-and-customer announcement of a Quebec deployment and of Lambda’s own latency and temperature measurements.
Sources
- Brian Sparks, “Lambda Boosts AI Factory Resiliency and Efficiency With NVIDIA Photonics,” NVIDIA blog, 16 November 2026
- Lambda, “Lambda announces the world’s first large-scale CPO deployment of NVIDIA Photonics switches”, 16 November 2026
- Lambda, “Lambda Announces World’s First Large-Scale CPO Deployment of NVIDIA Photonics Switches,” Business Wire, 16 November 2026
Get TSN’s biggest tech stories, explained.
News and insights across AI, tech, crypto and cybersecurity. Every source linked. Unsubscribe anytime.
Get the newsletterRelated stories
- Lambda Alabama Campus Pegged at $33.1bn Investment is the closest recent TSN account of Lambda’s wider campus build-out, and the right next stop for that company’s infrastructure file.
- UK Photonics Start-up Salience Labs Raises $125 Million to Scale Optical Switches for AI Data Centres
- PsiQuantum Turns Its Quantum Photonics Into an Optical Switch for AI
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X





