Lambda has deployed Nvidia's co-packaged optics technology in an AI cluster connecting more than 10,000 Blackwell Ultra GPUs, putting the architecture into production at a scale that tests its potential to reduce networking power demands.
As AI clusters grow, the networks connecting their accelerators consume more electricity, generate more heat and introduce additional failure points. In power-constrained data centers, that overhead competes directly with the capacity available for compute.
For operators, co-packaged optics (CPO) could free power for additional compute without expanding facility capacity. The tradeoff: integrated optical components may be harder to service or replace than conventional pluggable transceivers.
"When the facility is power-capped (and nearly all are), every watt the network uses is a watt that isn't going to compute," said Matt Kimball, vice president and principal analyst of data center technologies at Moor Insights & Strategy. "So network efficiency is no longer a networking problem – it's also a capacity and revenue problem."
Lambda estimates that a 128,000-GPU data center using conventional pluggable optics would require approximately 655,000 discrete transceiver modules across its switching fabric. Each consumes power, generates heat and represents another potential failure point.
Ron Westfall, vice president and practice lead for networking and infrastructure at HyperFrame Research, said the deployment represents an important architectural step as AI clusters grow. Networking infrastructure can become a constraint on power, thermal management and compute utilization, he said.
The deployment moves co-packaged optics beyond demonstrations and into a large production environment, but one installation does not establish a market. Operators must weigh potential efficiency and reliability gains against serviceability requirements, manufacturing challenges and tighter ties to a vendor's networking ecosystem.
Lambda's Quebec installation uses Nvidia Photonics switches to connect GB300 NVL72 systems. The switches integrate optical engines alongside switching silicon, shortening the electrical path between components and reducing the power and heat associated with conventional pluggable transceivers.
The systems have 144 ports running at 800 Gb/s, liquid cooling and 18 removable external light sources. Technicians can replace the light sources in the field, while the optical engines remain integrated into the switch package.
Lambda reported a 3.8% improvement in rack-scale latency and a 10.1% improvement in pod-scale latency compared with conventional pluggable optics, at the same throughput. Nvidia confirmed the figures represent network latency measured using a microbenchmark, not end-to-end application performance.
Networking power and compute capacity
Nvidia claims Photonics can deliver up to five times better power efficiency and up to 15 times higher reliability than conventional pluggable optics. The company also claims the architecture uses one-quarter as many lasers as conventional pluggable optics, reducing the number of components that could fail. An Nvidia spokesperson said the reliability figure reflects field-performance testing of signal-path reliability.
The power savings can be substantial at the component level. Nvidia said each CPO switch replaces 25-watt, 1.6-Tb/s transceivers with optical components consuming five watts for the same bandwidth. Across 72 transceivers, that amounts to a potential 1.44-kilowatt reduction per switch in optical-component power – not total switch power.
Lambda has separately modeled savings across larger installations. In a June technical analysis, the company estimated that a 10,368-GPU GB300 cluster could free up 658 kW by using co-packaged optics instead of conventional pluggable optics. Lambda said that would provide enough power for approximately 470 additional Blackwell Ultra GPUs.
The figure is a modeled cluster-level estimate, not a measurement from the Quebec deployment. Lambda's analysis also compared power consumption of approximately 3.95 kW per CPO switch with 7 kW for a conventional switch configuration. Those figures cover a broader switch-power comparison and are not directly comparable with Nvidia's optical-component estimate.
Kimball said reducing networking power can make more capacity available for revenue-generating compute, particularly in facilities where electricity availability limits expansion. He expects co-packaged optics to become the default architecture for the largest scale-out AI networks over the next several product generations, although technical and operational challenges will take time to resolve.
Serviceability remains an open question
Co-packaged optics change how operators maintain networking equipment. Conventional pluggable transceivers can be removed and replaced individually. With CPO, optical engines sit alongside switching silicon, making them part of the switch assembly.
The removable light sources in Lambda's deployment give technicians a way to service that portion of the system without replacing the entire switch. Nvidia said it and its ecosystem partners test components and assemble switches before shipment so equipment arrives in a known-good state. The company did not detail how operators would repair or replace an integrated optical engine if it failed.
Westfall cautioned that localized optical or thermal failures could have broader consequences for the switch assembly. He also cited manufacturing yields, supply-chain constraints and limited interoperability as potential obstacles to adoption.
"Operators must weigh these significant day-to-day energy and uptime benefits against stricter serviceability requirements and deeper architectural lock-in to Nvidia’s end-to-end networking stack," Westfall said.
Nvidia sees the technology as a way to improve network efficiency and reliability as AI infrastructure expands.
"In an AI factory, the network is a business accelerator. Every interruption stalls the job, and every watt counts," Gilad Shainer, senior vice president of networking at Nvidia, said in a statement. "Nvidia Photonics cuts network power, improves resiliency and delivers up to 15x higher reliability, so Lambda's customers get more productive GPU hours from their investment."
For Lambda, the deployment tests whether those gains can be realized in a large production environment.
"Networking has to keep pace with the demands of increasingly large AI clusters," Dave Ward, president of cloud services at Lambda, said in a statement. "Bringing co-packaged optics into a deployment of this scale is an important step toward building the efficient and resilient infrastructure needed to support the next generation of AI."
Shane Snider is a senior news writer at TechTarget, covering AI infrastructure, hyperscale data centers, cloud platforms, and the power and energy systems driving modern compute expansion. You can reach Shane at [email protected] or on LinkedIn.