Datacenter runs on green hydrogen and wastes zero water (www.theregister.com)

Lambda has deployed its Nvidia GB300 NVL72 system in an off‑grid ECL “bit barn” at the Mountain View MV1 campus that’s powered entirely by hydrogen fuel cells. The cells generate electricity by combining hydrogen with atmospheric oxygen (like spacecraft systems), producing water as a byproduct that ECL says is recycled for facility cooling—so the site claims zero net water consumption. The MV1 site can support up to ~1 MW of capacity, and Lambda has boosted its lease to 100% of the facility, but that still only accommodates a handful of GB300 racks given their power draw. The technical punch: each Supermicro GB300 NVL72 rack packs 72 Nvidia Blackwell Ultra accelerators, ~20 TB of HBM3e and roughly an exaFLOP of dense FP4 performance, but is rated at ~142 kW of power. A full eight‑rack “Superpod” would exceed 1 MW, highlighting power density and cooling as major constraints. The broader implication for AI/ML is twofold—hydrogen can enable off‑grid, high‑density AI sites and reduce fossil‑fuel emissions if the gas is truly “green” (electrolyzed using renewables), but carbon benefits hinge entirely on hydrogen sourcing and distribution. ECL’s MV1 is a prototype for much larger plans (a TerraSite campus targeting gigawatt scale), underscoring both the promise and scalability questions around hydrogen‑powered AI infrastructure.