# Oracle’s Project Jupiter Ditches Gas Turbines for Bloom Fuel Cells — retained record > Retained news record from news:datacenterknowledge. Canonical: https://highsignal.app/data/records/9a6baca2c68ca8f0 [← sources](https://highsignal.app/data) source: [news](https://highsignal.app/data/news) (news:datacenterknowledge) published 2026-04-29 12:54 ingested 2026-04-30 08:09 [original source ↗](https://www.datacenterknowledge.com/build-design/oracle-s-project-jupiter-ditches-gas-turbines-for-bloom-fuel-cells) ## extract Oracle’s Project Jupiter Ditches Gas Turbines for Bloom Fuel Cells The Bloom Energy deal shifts planned New Mexico AI campus to on-site generation. Oracle plans to replace gas turbines and diesel backup with a fuel-cell-based microgrid at its planned Project Jupiter data center campus in Doña Ana County, New Mexico, shifting the facility toward on-site generation for high-density AI workloads. The revised design consolidates the campus into a single microgrid sized to accommodate up to 2.45 GW of on-site capacity, marking another example of how grid constraints are driving the adoption of microgrids for high-density AI workloads. Project Jupiter is a planned AI data center campus spanning roughly 1,400 acres and expected to include four hyperscale buildings. The project has been described as a long-term investment that could reach up to $165 billion, with developers projecting about 4,000 construction jobs and roughly 1,500 ongoing roles. Oracle has said the system will operate without combustion and use minimal water for power generation, paired with closed-loop, non-evaporative cooling. On-site Power Replaces Conventional Backup In earlier plans, the campus included gas turbines and diesel generators. The current design eliminates those assets in favor of fuel cells and storage. According to Oracle, the system cuts nitrogen oxide emissions by roughly 92% and avoids the water use associated with conventional power generation. Oracle has indicated it expects to spend about $50 billion on AI infrastructure this fiscal year, and its agreement with Bloom Energy spans up to 2.8 GW of capacity, including 1.2 GW already under contract. Rob Gramlich, president of Grid Strategies, said grid limitations are prompting increased adoption of on-site generation. “Large data center operators still generally prefer grid power and use on-site generation primarily as backup,” he said. Bloom Energy fuel cells tackle grid limitations, providing reliable power for HPC data center operations. (Image: Bloom Energy) Power, Cooling, and Protection Converge Alex Cordovil, research director at Dell’Oro Group, said the shift reflects how AI infrastructure is being scaled. “The biggest contributor to extreme co-design across data center architecture is still rack density,” he said. “At a gigawatt scale, you have to extrapolate that same logic to the entire data center. Multiple complex systems are being pulled into the same physical envelope, and all of them need to be designed in tandem.” Prithpal Khajuria, segment leader for energy and utilities at Intel and a founding and steering committee member of the vPAC Alliance, said on-site generation changes where engineering challenges sit. In an off-grid configuration, operators avoid coordinating with utility protection systems, but they must still model and manage internal systems, including fuel cells, battery storage, and other electrical assets, to handle faults and load variation. Fully off-grid systems are costly, Khajuria noted, and many operators deploy a limited microgrid first, planning to connect to the grid within two to three years. Once connected, protection coordination becomes more complex, requiring models of both the microgrid and the utility grid, along with their interaction. Khajuria said most protection systems in use today are static, so operators will need systems that can adapt to changes in generation, storage, and load. Challenges Shift to Power and Cooling According to Cordovil, on-site generation does not change how compute and networking systems operate, but it does reshape facility design. “For networking and computing, on-site generation does not really change the picture yet. They are largely agnostic to whether the electron came from the grid or an onsite plant,” he said. “For the rest of the facility – both power and cooling – it changes the calculus completely.” At the multi-gigawatt scale, he added, power delivery and thermal density now rival networking as dominant constraints. “Networking topology has historically been the dominant constraint,” he said, “but at multi-gigawatt scale, power delivery and thermal density have caught up – and in some sites they are now the binding constraint.” Microgrids Expand Where Grid Limits Tighten Khajuria said microgrids give operators more control but do not eliminate risk. “A microgrid brings its own challenges, but it is manageable because everything is in the data center operators’ control,” he said. “[That] headache [is] pushed to the future as soon as grid connectivity comes into play.” Large campuses, Khajuria added, cannot rely on the grid as an immediate fallback if local generation fails. “They cannot flip the switch and ask the grid to provide power,” he said. Gramlich said integrating on-site generation with the grid is manageable from a system perspective. “We already have many large generators, loads, and hybrid storage-generation plants on the system which can be managed,” he said. However, Gramlich noted that on-site generation does not eliminate the need for grid infrastructure. “You still need grid connections and a robust backbone network to be ready when they do connect to the system,” he said. Cordovil said on-site generation is becoming more common in markets where grid access is constrained. “It becomes common where it has to, and remains situational where it can,” he said. He added that the use of fuel cells at this scale reflects a shift in siting priorities, with emissions, water, and community impact now part of the equation alongside cost and time to power. Fuel Cells Favor Stability Over Flexibility Fuel cells introduce operating constraints under variable AI workloads. “Fuel cells have very limited capacity to handle overloading,” Khajuria said. “They like ‘chill mode’ when the load is consuming power at a very steady rate.” To manage variability, operators must either size systems for peak demand or use batteries to absorb fluctuations. ## retained text No document body was retained for this record — only its metadata was stored.