RealityHackerOpen in RealityHacker ⇢
Compute · Networking & Interconnects · published 2026-11-16T00:00:00+00:00 · via Nvidia Blog

Lambda Enhances AI Infrastructure Resilience Using NVIDIA Photonics Switches

Image via Nvidia Blog
Image via Nvidia Blog

Lambda is deploying NVIDIA Photonics co-packaged optics (CPO) switches to connect over 10,000 GPUs across GB300 NVL72 racks, improving network efficiency and reliability. The technology reduces power consumption by up to five times and increases system resiliency by using fewer lasers and enabling cooler operation with a median optics temperature of 36°C. These upgrades support scalable AI training at large scale with lower latency and higher mean time between interruptions.

Expanded Detail

In November 2026, Lambda, a cloud infrastructure provider, deployed NVIDIA Photonics co-packaged optics (CPO) switches to interconnect over 10,000 GPUs across GB300 NVL72 racks in its AI factory. The deployment leverages 144-port switches operating at 800Gb/s per port, using a liquid-cooled system that maintains a median optics temperature of 36°C—significantly cooler than the 51°C typical of conventional pluggable optics. This shift enables up to five times better power efficiency and up to 15 times higher reliability by reducing the number of lasers used by fourfold and eliminating retimer chips. The co-packaged design shortens electrical signal paths from inches to millimeters, minimizing signal loss and improving stability. Each switch includes 18 replaceable external light sources, allowing field maintenance without replacing entire transceivers—a key advancement in system resiliency. These upgrades contribute to lower network latency (up to 10% improvement at pod scale) and longer mean time between interruptions (MTBI), critical for large-scale AI training and agentic inference workloads. Dave Ward, Lambda’s cloud services president, emphasized that this deployment marks a pivotal step toward supporting next-generation AI infrastructure demands.

The move follows growing industry pressure to scale AI compute while managing energy use and system reliability. As AI models grow in size and complexity, network bottlenecks have become a major constraint, with even minor disruptions stalling thousands of GPUs. Prior deployments relied on pluggable optics, which, despite being standard, introduced higher power consumption, greater heat output, and more failure points. Lambda’s adoption of CPO technology reflects a strategic shift toward integrated, efficient interconnects that align with broader trends in data center innovation. NVIDIA, as the developer of the Photonics platform, is positioning this technology as foundational for future AI infrastructure. What remains unresolved is how widely this approach will scale beyond Lambda’s deployment—whether other cloud providers will follow, and whether CPO will become a standard in next-gen AI clusters.

Context

Lambda’s adoption of NVIDIA Photonics could signal a broader industry pivot toward co-packaged optics as the standard for high-performance AI infrastructure. By drastically reducing power consumption and improving resiliency, this technology may help mitigate the environmental and economic costs of scaling AI compute, potentially influencing data center design and energy policy. The reduction in failure points and field-replaceable components could lower operational downtime, benefiting enterprises relying on continuous AI inference and training. A comparable precedent is the shift from traditional copper to optical interconnects in the early 2010s, which similarly transformed data center efficiency but required significant infrastructure retooling. Today’s CPO deployment may follow a similar trajectory—initially niche, then mainstream as performance and cost advantages become undeniable. However, widespread adoption depends on ecosystem maturity: whether switch vendors, OEMs, and software stacks will fully support CPO integration. The next phase to watch is whether other cloud providers replicate Lambda’s model, and how quickly standards bodies formalize CPO specifications—could accelerate industry-wide change in the coming years.

Expanded detail and Context are AI-generated analysis; the linked article remains the authoritative source.
Read the full article at Nvidia Blog →
Related stories
NVIDIA Vera Rubin Achieves Record MLPerf Results in Training and Agentic Inference · AI Accelerators
This summary is Al-enhanced to contain extended analysis and broader social context. The original is {NAME); the linked article is the authoritative source. Original headline: “Lambda Boosts AI Factory Resiliency and Efficiency With NVIDIA Photonics.” Browse more stories.