Oracle has abruptly changed the power strategy for its massive Project Jupiter AI data center campus in New Mexico, abandoning plans for a gas-turbine power plant after mounting opposition from regulators and local residents.
Instead, the company will rely on fuel-cell technology from Bloom Energy to generate electricity on-site, launching one of the largest and most ambitious experiments yet in alternative power systems for hyperscale AI infrastructure.
The change highlights a well-established megatrend challenging the AI industry: access to electricity has become a major constraint for AI expansion. In response, many operators are embracing what the industry calls BYOP (bring your own power), building dedicated generation systems directly at their data facilities rather than waiting years for grid interconnections.
Largest Fuel-Cell Deployment for a Data Center
Project Jupiter is tied to Oracle’s enormous cloud infrastructure partnership with OpenAI, part of a larger plan expected to involve hundreds of billions of dollars in AI-related investment. The massive New Mexico campus is projected to span roughly 1,400 acres and eventually support four hyperscale facilities designed for compute-intensive AI workloads.
While earlier plans called for natural gas turbines, those proposals generated spirited resistance from local activists concerned about air quality and water consumption in drought-prone Doña Ana County.
The New Mexico Environment Department received thousands of public comments regarding the project’s environmental impact, including concerns over groundwater usage and emissions. Additionally, both the Federal Energy Regulatory Commission and the New Mexico State Land Office rejected requests connected to a proposed gas pipeline intended to serve the site.
After Oracle withdrew permit applications tied to the gas plant, the company now plans to deploy Bloom Energy fuel cells as the backbone of an on-site microgrid capable of producing as much as 2.45 to 2.8 gigawatts of electricity. At that scale, the system would rank among the largest fuel-cell deployments ever attempted for a data center operation.
Unlike conventional combustion-based generation, Bloom’s solid oxide fuel cells generate electricity electrochemically using fuels such as natural gas or hydrogen, with the Project Jupiter plant expected to primarily use natural gas. Oracle claims the design will reduce nitrogen oxide emissions by roughly 92% compared with the prior turbine plan while sharply lowering water consumption through closed-loop, non-evaporative cooling systems.
Continued Criticism
The revised plan has not eliminated criticism. Environmental groups argue the fuel cells will continue to rely on natural gas and therefore remain tied to fossil fuel emissions. Local activists have also questioned why Oracle did not pursue a lower-emission strategy from the outset.
The scale of Project Jupiter illustrates how quickly infrastructure requirements are escalating. AI training and inference clusters now demand enormous amounts of electricity, cooling capacity, and networking throughput, forcing operators to rethink conventional data center architecture. As the AI buildout progresses, power delivery and thermal management have become just as important as compute hardware itself.
Fuel cells offer their own challenges. Unlike gas turbines, fuel-cell systems perform best under relatively stable loads, while AI workloads can fluctuate significantly. Data centers may need large-scale battery systems to smooth demand spikes and maintain reliability.
Oracle is taking on significant financial risk as it invests in AI infrastructure expansion. The company has secured more than $16 billion in financing for another large data center campus in Michigan as part of Stargate-related development efforts tied to OpenAI. For now, Oracle appears committed to rapid AI expansion regardless of the effort needed to balance environmental concerns, grid limitations, and enormous computational demand.
