Lambda runs hydrogen powered GB300 NVL72

This shows Lambda is moving from renting single GPU servers to operating full AI factory infrastructure. A GB300 NVL72 is not a normal cloud instance, it is a full liquid cooled rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs, built for giant training and inference jobs. Running that system at ECL on hydrogen power means Lambda is proving it can secure advanced Nvidia hardware, place it in production, and solve the power and cooling bottleneck that now matters almost as much as getting the chips themselves.1 2 3 4

• The practical point of hydrogen here is site speed and power density. Each deployed system gets 142 kW of compute power at ECL Mountain View, and the site runs off grid on hydrogen fuel cells, which lets Lambda bring dense AI capacity online without waiting for a conventional utility upgrade.2 4 5
• This puts Lambda in the same strategic lane as CoreWeave and Crusoe, where the product is no longer just GPU hours but guaranteed capacity inside purpose built data centers. Lambda had already shifted upmarket into multi year Nvidia and Microsoft deployments, while Crusoe differentiated through energy infrastructure and CoreWeave through hyperscaler scale.6 7 8
• The hardware itself signals the kind of customer Lambda is chasing. Nvidia describes GB300 NVL72 as a rack scale system for frontier AI workloads, with liquid cooling and tightly integrated networking. That is the gear used for training large models and serving heavy reasoning workloads, not casual developer experiments.1 3

The next phase of competition in GPU cloud is a race to turn power, land, cooling, and financing into usable clusters faster than rivals. If Lambda keeps pairing scarce Nvidia racks with unconventional energy sites, it can win customers that care less about brand name cloud and more about getting massive training capacity live first.2 6 7 8