The pipeline that would fuel Oracle's largest planned data center has slipped by nearly six months, and the delay underscores how land-use disputes can slow even the most ambitious artificial intelligence infrastructure projects.
Energy Transfer's Transwestern unit told federal regulators on Friday that the Green Chile Project now has an in-service date of 1 February 2027, against a previous target of 15 August. The filing with the Federal Energy Regulatory Commission marks the latest setback for Project Jupiter, a massive data center campus in Doña Ana County, New Mexico, that is expected to run on natural-gas-fueled Bloom Energy fuel cells and host AI infrastructure for OpenAI.
A data center on an unprecedented scale
Project Jupiter is not a typical data center development. It is planned at 2.5 gigawatts, with developers Stack and BorderPlex Digital Assets discussing investments of up to $165 billion across 1,400 acres and four buildings. The scale is so large that the natural gas pipeline intended to serve it would carry up to 400 million cubic feet per day, roughly 0.4% of all gas produced in the lower 48 states.
The numbers are difficult to overstate. A single gigawatt is enough to power around 750,000 homes, depending on the region. At 2.5 gigawatts, Project Jupiter would be among the largest data center campuses ever proposed. The involvement of Oracle as the tenant, with OpenAI as the underlying AI workload, suggests the facility is being designed for the next generation of large-scale machine learning models, which require enormous amounts of electricity and cooling.
The project is also part of a larger shift in how cloud companies are building capacity. Rather than relying solely on centralized cloud regions, hyperscalers and AI companies are increasingly demanding dedicated, highly dense data centers located close to power sources. This has created a new kind of competition, not just for chips and customers, but for land, water, and natural gas pipelines capable of delivering large volumes of fuel on demand.
An elected official in the way
Standing in the path of the pipeline is New Mexico land commissioner Stephanie Garcia Richard, an elected official who oversees state trust lands. She has refused to allow the pipeline to cross state trust land twice, first in March and again on 15 July. Her objections are not procedural. In her written decision, she argued that the project would benefit investors and developers but bring no significant benefits for state lands. She described the burden on New Mexico's water and natural resources as extreme.
On emissions, she was even more direct, saying that approving the pipeline would be “literally doubling down on dangerous emissions” at a time when the state should be slowing climate change rather than accelerating it. The commissioner's stance reflects a growing tension between the data center industry's demand for reliable power and the climate goals of the states where those data centers are built.
Water concerns in a dry state
Water is a particularly sensitive issue in New Mexico, which has been dealing with drought and the shrinking Rio Grande for decades. Data centers themselves use water for cooling, and gas extraction can also place pressure on local water supplies. Garcia Richard's objection highlights the fact that even though fuel cells are more efficient than traditional combustion turbines, the entire supply chain from gas well to power generation has environmental consequences.
The commissioner wrote that the project's impacts on water and natural resources would be profound, and that the economic benefits promised by the developers did not justify the long-term costs to the state. This is part of a broader pattern across the American West, where data center proposals are colliding with water scarcity, wildfire risk, and community concerns about industrial development.
Oracle argues time is of the essence
Oracle has made no secret of its urgency. In a May filing with the Federal Energy Regulatory Commission, the company said that “time is of the essence” and that any delay would jeopardize the broader objectives of Project Jupiter. The company has been borrowing heavily to build data centers as it races to challenge cloud leaders like Amazon Web Services, Microsoft Azure, and Google Cloud in the AI era.
Oracle's aggressive expansion strategy depends on having power in place when customers like OpenAI need it. A delay of nearly six months could ripple through the company's contracts and its ability to deliver AI services at scale. Investors reacted to the news on Friday by sending Oracle shares down about 4%, while Energy Transfer rose 1.4%, a sign that the market sees the delay as a negative for Oracle but not necessarily for the pipeline operator.
The fuel-cell compromise
Interestingly, the fuel cells at the heart of Project Jupiter were already a compromise. Bloom Energy's solid oxide fuel cells replaced the gas turbines and diesel generators that were originally planned. Oracle has since expanded its Bloom order to 2.8 gigawatts, suggesting that it sees the technology as a way to reduce emissions while still maintaining reliable, on-site power. But fuel cells still burn natural gas, and the industry as a whole continues to rely heavily on fossil fuels despite its public promises of sustainability.
Tech companies often market themselves as leaders in renewable energy, but their data centers are increasingly turning to natural gas to bridge the gap between grid capacity and the explosive growth in AI computing. Some companies have signed power purchase agreements with solar and wind farms, but the intermittency of renewables makes them insufficient for facilities that need to operate around the clock. Natural gas, while cleaner than coal, still produces carbon dioxide and methane emissions.
The choice of fuel cells also reflects a broader tension within the industry. On one hand, tech giants have pledged to reach net-zero emissions and are spending billions on clean energy projects. On the other hand, the immediate power demands of AI are so enormous that natural gas has become the most pragmatic option for many hyperscale projects. The Project Jupiter dispute brings this contradiction into sharp focus, forcing public officials to weigh private investment against public goods.
A wider pattern of local resistance
The Project Jupiter delay is also a story about land, not just capital. Money and computer chips can be acquired at speed, but a 17-mile right of way cannot. The pipeline route crosses state trust land, and the land commissioner has the power to veto it. In the broader data center industry, this is becoming an increasingly common obstacle. According to community groups and industry observers, around 500 towns in the United States have already blocked data center development in their jurisdictions, either through moratoriums, zoning restrictions, or outright opposition.
The reasons for local resistance vary. In some cases, it is about noise and visual impact. In others, it is about water use, electrical grid strain, or the environmental footprint of natural gas pipelines. There is also a growing concern that data centers create relatively few permanent jobs after construction is complete, while consuming a large share of local power and water resources. This has led to a backlash in places as different as Northern Virginia, Oregon, and New Mexico.
State trust lands are a particularly tricky hurdle. In New Mexico, these lands were set aside by Congress to generate revenue for public schools and other institutions. The land commissioner is required to manage them in a way that benefits the trust beneficiaries, not necessarily to maximize private commercial returns. That means a project must show a clear and direct benefit to public schools, universities, and other state institutions. Garcia Richard determined that the Green Chile pipeline did not meet that standard, even though the data center itself could eventually generate tax revenue.
The delay is also a reminder of the physical limits of the AI boom. Every large language model, every autonomous vehicle system, and every generative AI application depends on data centers that require unprecedented amounts of electricity. The companies building that infrastructure are discovering that securing permits for power generation and pipelines is often more difficult than building the facilities themselves. Decades-old laws governing land use, environmental review, and public utilities are now becoming critical bottlenecks for the world's most valuable technology companies.
The Green Chile pipeline is now scheduled to go into service on 1 February 2027, nearly six months later than originally planned. The delay may be only the beginning. The land commissioner has rejected the route twice, and it is unclear whether Energy Transfer and the developers can find an alternative path or persuade her to change her mind. The state's trust land board, which oversees the commissioner's office, could also become involved.
For Oracle, the stakes are enormous. The company has staked its future on being able to build hyperscale data centers quickly and cost-effectively. But the physical reality of land, water, and local politics is proving harder to overcome than the technical challenges of chip supply or server deployment. The delay of the Green Chile pipeline is a reminder that in the age of AI, the most valuable commodity is not compute but consent.