Hydrogen power is getting a serious test run in Silicon Valley. ECL and Swedish fuel-cell maker PowerCell have signed a commercial partnership to install hydrogen fuel-cell power systems across U.S. data centers, with the first gear headed to ECL’s new 35 MW CSC-1 campus in Santa Clara. The deal bundles a firm purchase order with a broader memorandum of understanding that targets hundreds of megawatts of future hydrogen capacity in ECL’s portfolio. ECL’s Mountain View pilot, MV-1, which has already been running on fuel cells, will supply the real-world data that shapes the broader rollout.
As reported by Data Center Dynamics, the agreement includes a firm purchase order for PowerCell PS190 fuel-cell systems plus a non-binding memorandum of understanding for roughly 300 MW of additional hydrogen capacity. PowerCell CEO Richard Berkling told the outlet the move sends a clear signal that hydrogen-powered AI data centers are moving from first-of-kind toward industrial scale. The partners say that pairing a concrete order with a larger MoU is meant to speed behind-the-meter deployments in markets where utility interconnection is slow or constrained.
First installs in Santa Clara
ECL plans to place the first containerized PowerCell units inside its FlexGrid microgrid at CSC-1, a 35 MW Santa Clara campus that blends grid electricity, natural gas and battery storage to keep capacity online, according to the company’s release on Businesswire. ECL pitches CSC-1 as a fast-build, modular site tuned for AI workloads, with power that can ramp in smaller steps instead of waiting months or years for utility upgrades. The company also says by-product water from hydrogen generation will be reused for cooling at the site to reduce outside water demand.
What the PS190 can do
PowerCell markets the unit as the Power System 190, or PS190, and in late 2025 said it had begun taking orders for the model as part of a new power-generation portfolio, per a PowerCell press release. The company describes the PS190 as a modular, containerized fuel-cell unit built for both mobile and stationary use. Data Center Dynamics reports that in ECL’s architecture, each containerized PS190 installation will deliver roughly 225 kW and can be scaled toward megawatt levels to support dense AI racks.
Why operators are buying in
Behind-the-meter fuel cells give data center operators a way to sidestep long utility interconnection queues while providing continuous, dispatchable power for high-density AI workloads. S&P Global forecasts that U.S. data-center electricity demand could climb from about 61.8 GW in 2025 to roughly 134.4 GW by 2030, a gap that is pushing interest in distributed firm power solutions. Hydrogen fuel cells promise lower operational emissions than diesel and a potential route to 24/7 low-carbon computing in places where grid upgrades are slow or simply not on offer.
From Mountain View to Texas
ECL’s MV-1 pilot in Mountain View has been using fuel-cell generation as a primary power source since mid-2025, and that operating history helped drive the company’s choice of PowerCell and Bosch as partners, according to industry coverage. Data Center Frontier also points to Lambda as an early tenant tied to ECL’s broader TerraSite-TX1 plans, where the company is targeting a 1 GW hydrogen-powered campus in Texas. ECL says lessons learned from MV-1 will be used to de-risk and accelerate larger deployments.
Manufacturing and the scaling challenge
PowerCell will lean on manufacturing support from Bosch, which has been described in company filings and releases as a major investor and manufacturing partner, to help industrialize the PS190 supply chain, according to PowerCell communications. Moving from pilot-scale megawatts to the multi-hundred-megawatt range envisioned in the MoU will still depend on reliable hydrogen supply, storage, permitting and local integration work, so the logistical and regulatory tail is likely to be long. Even so, the ECL and PowerCell agreement stands out as one of the clearest commercial steps so far toward making hydrogen fuel cells a practical option for Bay Area AI compute.
