On September 24, 2025, Lambda teamed up with ECL and Supermicro to fireside up the very first production-grade NVIDIA GB300 NVL72 AI programs at ECL’s off-grid “MV1” campus in Mountain View, California. Working fully on hydrogen gasoline cells, this modular information middle isn’t only a high-performance powerhouse—it’s a showcase for true zero-emission know-how and nil exterior water use. In a world the place huge basis fashions push conventional energy and cooling to the brink, MV1 proves sustainable AI compute isn’t a pipe dream—it’s right here, and it really works.
Reimagining Hydrogen-Powered AI Compute
By tapping into hydrogen manufacturing through gasoline cells, ECL’s MV1 stays fully off the grid and out of the municipal water loop. Even higher, the cooling circuit will get a second life with the very water produced throughout energy technology—captured, purified, and recycled again into the system. ECL claims this might web a adverse native water footprint (pending unbiased validation). That’s the sort of real-world influence that makes inexperienced hydrogen and cutting-edge cooling applied sciences extra than simply buzzwords.
- First-ever production-grade NVIDIA GB300 NVL72 AI programs powered fully by inexperienced hydrogen
- Every Supermicro cupboard pulls 142 kW of compute muscle
- Plug-and-play rack integration in simply two hours
- No diesel turbines buzzing within the background
- Zero exterior water draw because of recycled byproduct water
Historic Context
Again within the early 2010s, modular information facilities began showing as a nimble, cost-effective different to conventional builds. On the similar time, hydrogen gasoline cells had been primarily relegated to telecom backup energy—temporary stints, nothing extra. Then, in mid-2023, ECL demoed a 5 MW inexperienced hydrogen module in Mountain View that ran 24/7 with out touching the grid. MV1 takes that proof-of-concept an enormous leap ahead by pairing hydrogen power with high-density GPU compute for the primary time at manufacturing scale.
MV1 isn’t simply flexing compute energy; it’s rewriting the playbook for information facilities in distant or resource-constrained websites. By ditching diesel turbines and mechanical chillers in favor of gasoline cells and closed-loop liquid cooling, it tackles each carbon emissions and water utilization head-on. With regulators and clients demanding greener setups, this sort of forward-looking design exhibits how hydrogen infrastructure can shatter previous limits on energy density.
Technical Below the Hood
Inside every cupboard, you’ll discover 72 NVIDIA GB300 GPUs speaking to 1 one other over NVLink for blistering bandwidth and rock-solid low latency. A direct-to-chip liquid cooling system—managed by centralized distribution models (CDUs)—retains temperatures in test. Right here’s the kicker: that coolant is definitely the water byproduct from changing hydrogen into electrical energy. Speak about a closed loop for each energy and thermal administration.
Every NVL72 rack is constructed with eight liquid-cooled trays, 9 GPUs apiece, all tuned for distributed coaching frameworks like PyTorch and Horovod. The system even recycles waste warmth from the gasoline cell stack, pushing PUE beneath 1.05—on par with essentially the most environment friendly hyperscalers.
- Hydrogen gasoline cells convert H₂ into electrical energy, with pure water as the one byproduct
- CDUs route that water by means of microchannel chilly plates hooked up on to GPU dies
- Superior warmth switch cuts total cooling energy, boosting effectivity
- Racks obtain full operation inside two hours, slashing deployment time
Strategic Implications
This landmark rollout underscores Lambda’s imaginative and prescient of gigawatt-scale “AI factories.” By doubling its footprint at ECL’s Mountain View web site, Lambda is staking its declare on hydrogen’s future in AI. Going off-grid insulates them from risky electrical energy costs and brownout dangers—big advantages as basis fashions demand ever extra compute. And that two-hour rack integration? It means AI startups and analysis labs can scale on demand with out a huge set up invoice. With energy prices climbing in tech hubs like Silicon Valley, MV1’s hydrogen-based economics are drawing severe curiosity from sustainability-focused traders. Up subsequent: TerraSite-TX1, a multi-gigawatt campus close to Houston the place Lambda plans one other hydrogen-fueled showcase. As ESG mandates tighten, MV1’s blueprint might turn out to be the gold customary for clear, high-performance compute.
Firm Context
Based in 2012 by a bunch of hands-on AI engineers, Lambda has turn out to be a go-to for scalable GPU cloud compute and colocation—its “Superintelligence Cloud.” ECL, led by CEO Yuval Bachar and backed by Hyperwise Ventures and Molex, focuses on off-grid, modular information facilities fueled by inexperienced hydrogen. Supermicro, headquartered in San Jose, builds the liquid-cooled racks that home NVIDIA’s highly effective GPUs. Collectively, they’ve woven compute, power, and {hardware} experience right into a seamless answer, letting purchasers faucet into devoted inexperienced compute with out constructing their very own mini energy vegetation—shortening time to market from years to months.
Environmental and Market Influence
Environmental: MV1’s zero-emission profile wipes out Scope 2 emissions, and by returning purified byproduct water to native aquifers (pending audit), it might even assist recharge groundwater.
Industrial: Demonstrates that off-grid, hydrogen-based information facilities can rival grid-tied amenities, reducing the necessity for brand new substations and transmission upgrades.
Market: Spurs funding in hydrogen infrastructure and superior cooling, opening contemporary funding channels for inexperienced AI ventures.
Technical: Validates that liquid-cooled GPU clusters can run reliably on modular hydrogen energy vegetation, paving the way in which for deployments in distant or resource-scarce environments.
Societal: Gives policymakers a dwelling instance of how compute-intensive industries can hit aggressive decarbonization targets, bolstering assist for hydrogen infrastructure packages.
Wanting Forward
MV1 resides proof that hydrogen-powered AI compute works—and the true take a look at comes when TerraSite-TX1 rolls out in Houston. If this momentum holds, we might see gigawatt-scale, hydrogen-fueled AI campuses throughout North America by late 2027. Advances in electrolysis—whether or not tied to renewables or next-gen feedstocks—might be key to scaling these zero-emission know-how hubs. As hydrogen gasoline cells proceed to mature, initiatives like MV1 will set the bar for decarbonizing high-performance compute at scale.
About ECL
Based by veterans from Fb, Microsoft, LinkedIn, Cisco, HPE, and Bloom Vitality, and led by CEO Yuval Bachar, ECL focuses on off-grid, modular, hydrogen-powered information facilities. Backed by Molex and Hyperwise Ventures, the MV1 pilot in Mountain View is ECL’s first full-scale proof-of-concept for zero-emission AI compute amenities.
