The off-grid, hydrogen-powered, modular data center company. PUE under 1.1. Zero emissions. Water as a byproduct.
No substation. No utility tie-in. No diesel generators behind a chain-link fence. The building exhales pure water and a faint warmth. Inside, racks of servers do the same kind of work they do everywhere else in Silicon Valley - except this room is, by any reasonable definition, off the grid.
This is ECL's MV1. It is the closest thing the data center industry has to a working answer for what comes after the current power crisis. ECL - short for EdgeCloudLink - is the company that built it, leased it (to cloud-bare-metal provider Cato Digital), and is now trying to do the same thing at the scale of a gigawatt outside Houston.
The pitch is short enough to fit on a napkin. Hydrogen fuel cells produce electricity and water. Use the electricity to run the servers. Use the water to cool them. Return whatever water is left to the surrounding community. Then build the whole thing in pre-assembled modular blocks so the next site can ship in months instead of years.
Around 2022 a few uncomfortable graphs started circulating among utility planners. AI training was projected to push hyperscale data center demand past the limits of regional grids in places like Northern Virginia, Dublin and Santa Clara. New sites were being told to wait five, sometimes seven years for an interconnect. New transmission lines were being told to wait longer.
The industry's usual answer to a power problem has been to find cheaper power, build closer to it, and run diesel generators when the grid burps. Yuval Bachar, who had spent roughly three decades inside that answer at Cisco, Juniper, Facebook, LinkedIn and Microsoft, thought the answer needed a rewrite. Diesel is filthy. Waiting five years for a substation is unacceptable. And the cooling water bill, in much of the world, is starting to become a political problem.
Field note. In 2023 a single Northern Virginia county told developers no new grid connections would be approved for new data centers until further notice. The phrase "further notice" is doing a lot of work in that sentence.
Yuval Bachar is ECL's co-founder and CEO. His resume is the kind that makes investors sit up straight: data center leadership at Microsoft, LinkedIn, Facebook (where he worked on the Open Compute Project), Cisco and Juniper. He also co-founded Open19, an open hardware standard adopted by enterprise users who liked the OCP idea but did not have hyperscale budgets.
His co-founder, Rajesh Gopinath, runs engineering. The two of them put together a thesis that reads, in retrospect, like an obvious arbitrage. Hydrogen fuel cells were getting cheaper and quieter. Modular construction worked. Air cooling had been pushed about as far as physics allowed. The cleverness was in stitching those three commodities into a single product nobody had bothered to package.
ECL incorporated in 2020. The seed round, $7M from Molex and Hyperwise Ventures, closed in early 2023. A $10M Series A followed in mid-2024, led again by Hyperwise. That is roughly $17M of capital to ship a working hydrogen-powered, off-grid, modular data center - a small number for the size of the swing.
Customers do not buy a building from ECL. They buy a block. One block is somewhere between one and two megawatts of compute capacity, pre-engineered, factory-built, and dropped into a configured site. Stack the blocks and you have a campus. The architecture is intentionally boring, in the way the shipping container was once boring.
1-2 MW modular block, the unit you stack to get a site. Built-to-order, shipped pre-assembled.
Runs on green hydrogen, natural gas, or grid - whichever combination the site needs, with seamless switching.
Cooling loops use the pure water generated by the fuel cells themselves. No municipal draw.
Predictive modeling, PUE monitoring, capacity planning across disaggregated sites.
MV1 is fully leased. The first tenant is Cato Digital, a cloud bare-metal provider that needed dense GPU capacity faster than the local grid was willing to provide it. That single contract retired the question of whether anybody would actually rent a hydrogen-powered data center. They will.
The harder test is the Houston announcement. ECL and Lambda - the AI cloud company - are jointly developing what they describe as a one-gigawatt, off-grid, hydrogen-powered "AI Factory" on six hundred acres outside the city. A gigawatt is roughly the output of a mid-sized nuclear reactor. The site is designed to host successive generations of NVIDIA GPU systems without ever connecting to ERCOT.
In 2025 the company won three categories at the Global Sustainability Awards - Net Zero, Project of the Year, and Startup. Awards are not customers, but for a company selling infrastructure to risk-averse buyers, awards are a useful proxy for credibility.
ECL's stated mission is to deliver community-integrated, flex-grid, modular AI data center solutions with autonomous, sustainable power and minimal environmental impact. Read carefully, the operative word is "community-integrated." Most data centers, regardless of efficiency, arrive in a town as a net draw on local resources. ECL's design intentionally returns clean water to the host community as a byproduct of operations.
It is a marketing line, of course. It is also a real engineering choice. The hydrogen fuel cells produce more pure water than the cooling loops require. The remainder has to go somewhere. ECL has chosen to make that "somewhere" the local utility, with appropriate paperwork.
Footnote for the cynics. The water-give-back program is small in absolute terms. The point is that the sign of the number is positive. For an industry that has spent two decades arguing about its water bill, that is not a trivial reframe.
If you spend a few hours reading earnings transcripts from the hyperscalers, you notice a quiet pivot. The bottleneck on AI growth is no longer GPUs. It is megawatts. It is interconnect queues. It is sub-stations that have a five-year backlog. It is, increasingly, water.
ECL is one of perhaps a dozen companies that have noticed the bottleneck has moved and have decided to sell shovels in the new place. The bet is not that hydrogen will win every site. It will not. The bet is that for a meaningful slice of new AI deployments, the right answer is a power plant in a box, shipped to a site the developer was already going to acquire, with no transmission queue and no community water fight.
That is not a small market. It is also not a market with a lot of competing answers. Modular nuclear is on the horizon but not on the loading dock. Behind-the-meter gas turbines work but bring their own permitting fight. Hydrogen, today, is the only one that produces drinkable water on the way out of the stack.
If you stand outside MV1 long enough, the building does something quietly subversive. It works. The racks heat up. The fuel cells answer. The water loop closes. Nothing in the sequence touches the grid. The only audible difference between this data center and any other data center on the same street is what it isn't doing.
ECL is not a finished company. The Houston campus is not yet built. The competitive set is moving. Hydrogen prices are volatile. The team is still small for the size of its ambitions. None of this is a secret to anybody involved.
But the building is real. The tenant is paying. The water is leaving the site cleaner than it arrived. For an industry that has gotten used to announcing things, ECL has the rarer privilege of running one.