THE CRUNCH
Cloud provider Lambda is using NVIDIA Photonics co-packaged optics switches to connect NVIDIA GB300 NVL72 racks in a cluster of more than 10,000 GPUs, reporting lower power draw, higher resiliency and longer mean time between interruptions at AI factory scale. Co-packaged optics place the optical engines that convert electrical signals into light on the same package as the switch chip, shortening the electrical path from inches to millimetres and removing the need for retimer chips.
In Lambda's tests, network latency dropped by about 4% at rack scale and by up to about 10% at pod scale, while throughput matched conventional pluggable optics across rack, scalable unit and pod configurations. NVIDIA claims the switches deliver up to 5x better power efficiency and up to 15x higher reliability than pluggable optics, and use 4x fewer lasers.
Each switch has 144 ports running at 800Gb/s, with liquid cooling carrying heat away from the chip and optical engines. Lambda reports a median optics temperature of 36°C, against 51°C for conventional pluggable optics, alongside more stable signal integrity.
The resiliency gains matter because in a large training job a single interrupted link can stall thousands of GPUs, and in agentic inference one disruption can cascade latency across many agents. Each switch also carries 18 removable external light sources that technicians can replace in the field, whereas a failed laser in traditional pluggable optics means replacing the entire module.
Side by side
Co-packaged optics vs pluggable optics
| Co-packaged optics | Pluggable optics | Difference | |
|---|---|---|---|
| Median optics temperature | 36°C | 51°C | +42% |
| Power efficiency | ≤5x better | — | – |
| Reliability | ≤15x higher | — | – |
| Lasers used | 4x fewer | — | – |
Details from NVIDIA Blog, as quoted in this brief.


