Oracle has scrapped plans for natural gas turbines and diesel generators at its enormous New Mexico AI campus, replacing them with up to 2.45 gigawatts of Bloom Energy solid-oxide fuel cells in what would rank as one of the largest fuel cell deployments ever attempted. The swap, announced in May 2025 by Oracle, the BorderPlex Alliance, and Bloom Energy, centers on Project Jupiter, a single microgrid campus near the U.S.-Mexico border built to run artificial intelligence workloads at scale.
If fully built, the installation would nearly double Bloom Energy’s entire lifetime deployment history. The company reported roughly 1.3 GW of cumulative installations through its most recent annual filing. Delivering 2.45 GW to a single customer, on a single site, represents an industrial challenge with few precedents in the fuel cell sector.
Why fuel cells instead of turbines
Solid-oxide fuel cells generate electricity by running natural gas or hydrogen through an electrochemical reaction rather than burning it. That distinction matters for two reasons in southern New Mexico.
First, the partners claim the fuel cells will produce approximately 92% less nitrogen oxide than the combustion equipment they replace. NOx is a precursor to smog and a trigger for respiratory illness, and the Chihuahuan Desert airshed already contends with cross-border pollution from the El Paso-Juarez metro area. While the 92% figure comes from the companies’ own joint announcement and has not been independently audited, it aligns with U.S. Department of Energy data showing that solid-oxide fuel cells emit a fraction of the NOx produced by gas turbines of comparable output.
Second, fuel cells sidestep the massive water demands of conventional thermal generation. Combined-cycle gas plants typically consume between 0.5 and 0.7 gallons of water per kilowatt-hour for cooling. At 2.45 GW, that would translate into billions of gallons annually. The companies describe the fuel cells’ water needs as “negligible,” a claim consistent with the technology’s lack of a steam cooling cycle. In a state where the Rio Grande regularly fails to meet downstream delivery obligations and water rights litigation has dragged on for decades, avoiding a new industrial draw on the basin removes a major permitting obstacle.
The microgrid model
Rather than plugging into the regional utility grid and competing for transmission capacity, Oracle and Bloom Energy are designing Project Jupiter as a self-contained power system sized to the campus’s AI demand. The approach reflects a broader shift among hyperscale data center operators, who increasingly view on-site generation as the fastest path to securing reliable power for machine learning workloads that can spike unpredictably.
Microsoft has pursued dedicated nuclear capacity through deals with Constellation Energy. Amazon has signed large-scale solar and wind agreements. Oracle’s bet on fuel cells occupies a different niche: generation that can be manufactured in a factory, shipped to a site, and stacked modularly without the years-long permitting cycles that slow nuclear and large renewable projects. Whether that speed advantage holds at 2.45 GW remains to be seen.
What the announcement leaves out
Several critical details are missing from the public record, and they will shape whether Project Jupiter lives up to its billing.
Timeline: Neither Oracle nor Bloom Energy has disclosed when the first fuel cells will arrive on site, when the campus will reach full capacity, or what interim power sources will keep operations running during construction. Manufacturing 2.45 GW of solid-oxide units would be a historic production run for Bloom, and supply-chain constraints on specialty ceramics and other cell components could stretch delivery schedules.
Cost: The announcement includes no contract value, no per-kilowatt price, and no comparison to what the original gas-and-diesel plan would have cost. Without those figures, it is impossible to judge whether Oracle is paying a premium for cleaner power or whether fuel cells have reached cost parity with combustion alternatives at this scale.
Fuel source: Solid-oxide fuel cells can run on pipeline natural gas, biogas, or hydrogen, and the carbon footprint of the finished campus depends heavily on which feedstock Oracle chooses. Running on unabated natural gas would still produce CO2, albeit less than combustion turbines per unit of electricity. Running on green hydrogen would slash carbon emissions further but would require a hydrogen supply chain that does not yet exist in the region at anything close to this scale.
Regulatory review: New Mexico’s Environment Department and Office of the State Engineer would typically evaluate air emissions and water appropriation for a project of this magnitude. No public filings from either agency have surfaced. Until permits and environmental impact assessments are on the record, the emissions and water claims remain company projections, not regulatory findings.
The broader Bloom-Oracle partnership
Alongside the Project Jupiter details, the two companies disclosed that their overall relationship now targets up to 2.8 GW of fuel cell capacity. Simple math leaves only 350 megawatts for every other Oracle site combined, suggesting that either additional deals have not been finalized or that New Mexico is, for now, the overwhelming focus of the partnership. Whether the 2.8 GW figure represents firm orders or a longer-term ceiling is unclear.
For Bloom Energy, the commercial stakes are enormous. The company’s stock has historically been sensitive to large contract announcements, and a deal of this size could redefine its revenue trajectory. For Oracle, the calculus is different: securing enough power, fast enough, to keep pace with AI infrastructure demand that shows no sign of plateauing.
What comes next for Project Jupiter
The gap between announcement and operation will be the real test. Regulatory filings, construction milestones, and independent emissions monitoring will determine whether Project Jupiter delivers on its environmental promises or whether the numbers announced in May 2025 were aspirational targets that shift as engineering realities set in. In the meantime, the deal signals that the race to power AI is pushing tech companies toward energy technologies that, even a few years ago, few would have considered at gigawatt scale.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
