Permitting Delays Put Oracle’s $165 Billion Project on Notice

Data Centers & Energy By EnergyInnoCore Editor · Published Oct 8, 2026 · Updated Oct 9, 2026 · 13 min read

On Thursday, September 24, a two word legal phrase moved billions of dollars of market value in a single morning.

Oracle sent a force majeure notice on Project Jupiter, the massive AI data center campus in Doña Ana County, New Mexico, that is expected to serve OpenAI as part of the broader Stargate buildout. Within hours, Oracle shares were down roughly 3 to 4 percent, Bloom Energy fell more than 6 percent at one point before closing about 5 percent lower, and Blue Owl Capital, whose data center unit is developing the campus, slipped as well.

Force majeure usually brings to mind hurricanes, wars, or pandemics. So what catastrophe hit Project Jupiter?

The answer is far less dramatic, and in many ways far more important for our industry. It was permitting.

More specifically, a 17 mile gas pipeline that could not get a right of way across state land, and an air quality permit for the fuel cells still working its way through hearings and courtrooms. No storm. No war. Just regulators saying no, or at least not yet.

In a previous article on fuel cells for Behind the Meter (BTM) power, I used Project Jupiter as an example of why a fuel cell does not automatically mean no air permit. I did not expect it to become front page financial news a few weeks later. So let's look into what actually happened, and why it matters well beyond the New Mexico desert.

What Happened on September 24

According to reports first published by Bloomberg, Oracle sent the notice to Stack Infrastructure, the data center development arm of Blue Owl Capital. Stack is building the campus, and Oracle is the tenant. Oracle, in turn, plans to provide the computing capacity to OpenAI.

The purpose of the notice is fairly narrow. Oracle is protecting itself from payment obligations, essentially rent and related commitments, if the facility is not delivered on its planned 2028 schedule because of events outside Oracle's control. The event cited is the delayed gas supply for the on site power plant.

Oracle was quick to calm the waters. The company said that Project Jupiter "remains on our planned schedule" and that it is "fully committed to New Mexico." Blue Owl's position, as reported, is that the notice does not change the parties' financial commitments.

So nobody is walking away. But in project finance, a force majeure notice is never just a formality. It signals that the contract's risk allocation is now in play, and every lender and investor reads it carefully.

What Force Majeure Actually Means Here

The developer, in this case Stack backed by Blue Owl, raises the capital and builds the facility. Reports indicate Blue Owl committed around $3 billion of equity and that roughly 20 banks provided an $18 billion construction loan. The tenant, Oracle, signs a long term lease. That lease, with a creditworthy hyperscaler on the other side, is what makes the entire capital stack financeable. Lenders are essentially lending against Oracle's promise to pay.

Now imagine the building is ready, but there is no power. Should Oracle still pay rent on a building it cannot use?

A force majeure clause is designed to answer exactly that question. If an event beyond the parties' reasonable control prevents performance, obligations can be suspended or deferred. Oracle's notice is an early flag saying, in effect, that if power arrives late because of the permitting situation, the rent clock should arrive late too.

For Oracle, this is prudent risk management. For Blue Owl and its lenders, it is a problem, because the lower returns during construction stretch out longer, and the higher returns after completion get pushed further into the future. Reports suggest that some of the related debt had already been trading at stressed levels before the notice. That is why the market reaction was not limited to Oracle.

The Real Story: A 17 Mile Pipeline

Project Jupiter was designed around on site power. The campus would not wait years for a utility transmission connection. Instead, it would bring its own generation. That generation needs fuel, and the fuel is natural gas.

The gas was supposed to come through the Green Chile project, a lateral pipeline proposed by Energy Transfer through its Transwestern Pipeline subsidiary. The line would run roughly 17 miles from the El Paso area to the campus, using a 24 inch pipe capable of delivering about 400 million cubic feet of gas per day. The estimated cost was around $60 million. For a project of this scale, that is a rounding error.

The route crosses federal land, private land, and New Mexico state trust land. And that mix of land ownership is exactly where the trouble began.

On the federal side, things moved remarkably fast. In May, the Bureau of Land Management's Las Cruces office approved the federal portion of the route using emergency permitting procedures adopted by the Department of the Interior in 2025. A review that would normally take about a year was compressed into roughly 14 days.

The state side went very differently.

New Mexico's Commissioner of Public Lands, Stephanie Garcia Richard, denied the right of way across state trust land in March 2026. Energy Transfer asked for reconsideration. On July 15, she denied it again.

Her reasoning is worth reading closely. She argued that the pipeline would generate only about $32,000 in one time payments plus roughly $217,000 over five years for the state land trust, a trivial benefit compared with the burden. She pointed to an estimate of about 10.1 million tons of greenhouse gas emissions per year from the power plant it would supply, a figure larger than the annual emissions of the Albuquerque metro area. And she described the burden the project would place on New Mexico's water and natural resources as extreme.

In other words, a state land office, an agency most data center developers would never think to put on their critical path, effectively stopped the fuel supply for a 2.45 GW campus.

The consequence showed up in Energy Transfer's schedule. The pipeline's target in service date moved from August 15, 2026, to February 1, 2027, as the company works on an alternative approach to the route. That is roughly six months of slippage on the one piece of infrastructure the entire power plant depends on.

The Second Front: The Air Quality Permit

The pipeline is only half of the story.

In April 2026, Oracle, BorderPlex Digital Assets, and Bloom Energy announced that Project Jupiter would replace its earlier plan of gas turbines and diesel generators with up to 2.45 GW of Bloom solid oxide fuel cells. The partners said the change would cut NOx emissions by about 92 percent compared with the turbine design and use negligible water. Oracle also committed $50 million to repair and upgrade local water systems.

On paper, this was a smart move. Fuel cells produce far fewer conventional air pollutants than combustion, which in several jurisdictions means a simpler permitting path.

But New Mexico still requires an air quality permit from the New Mexico Environment Department for the fuel cell plant. And that process has been anything but smooth.

The public hearing on the permit was scheduled, rescheduled, and then moved up. In August, two groups, New Energy Economy and the Center for Biological Diversity, asked the New Mexico Supreme Court to intervene, one petition concerning the air permit and another concerning the use of an emergency water well for construction. The Court paused the proceedings. On September 17, it unanimously rejected both challenges and allowed the process to resume. However, the hearing officer assigned to the case had recused himself in August, and as of mid September no replacement had been named and no new schedule had been set. Reports indicate the permit needs to be issued before late November to keep the plan on track.

A permit hearing that has not been rescheduled cannot produce a permit.

Why Fuel Cells Did Not Escape Permitting

The case for fuel cells in BTM power rests heavily on speed, and I still believe that argument is valid.

But Project Jupiter shows two limits very clearly.

First, lower emissions are not zero emissions. A fuel cell running on natural gas still produces carbon dioxide. At a scale of 2.45 GW running around the clock, the absolute numbers become very large, large enough for opponents and regulators to compare them with entire cities. The technology may be cleaner per megawatt hour, but a gigawatt scale campus is judged on its total footprint.

Second, the fuel cell does not stand alone. It needs gas, and gas needs a pipeline. The permitting critical path therefore includes not just the power plant, but every piece of infrastructure that feeds it: the pipeline right of way, federal pipeline approvals, water for construction, and local land use decisions. A developer can pick the most permit friendly generation technology available and still be stopped by a land office that controls a few miles of the route.

This is the lesson I keep coming back to. Technology choice and permitting strategy cannot be separated. They have to be designed together, at the site selection stage, with the entire supply chain of the power source mapped out.

How a Permitting Delay Ripples Through Development

Why does a six month pipeline delay create this much noise? Because in AI data center development, delays do not add up. They multiply.

Consider the chain of events. The pipeline slips six months. The fuel cells cannot be commissioned without gas, so the power plant slips. Without firm power, the data halls cannot be energized, so the GPUs cannot be installed and tested. Without operating capacity, Oracle cannot deliver compute to OpenAI. And without delivered compute, the revenue that justifies the entire investment moves to the right. Reports have described the potential delay to the overall project as up to about a year.

Each link carries its own cost. Construction loans keep accruing interest. Fuel cell orders have to be deferred or redirected. GPU generations keep turning over, so hardware bought for a 2028 launch may be a generation behind in 2029. And the contractual chain, from lender to developer to tenant to end customer, strains as each party looks for protection. That is exactly what Oracle's notice did.

The Market Reaction: Oracle, Bloom, and Blue Owl

The share price reactions tell us what investors are really worried about.

For Oracle, the concern is not one campus. It is the balance sheet behind its entire AI strategy. Oracle reported $28.5 billion of capital expenditures in its latest quarter, up from $8.5 billion a year earlier, with negative free cash flow of around $5 billion. Its remaining performance obligations are enormous, reported at around $664 billion, but those contracts only turn into cash when capacity is delivered. A KeyBanc analyst described Oracle as being in "the most tenuous of cash positions of all the major hyperscalers." When a company in that position signals a possible delay, investors naturally ask whether it is the first of many.

For Bloom Energy, the worry is timing. Project Jupiter is one of the largest fuel cell deployments ever announced, and the stock had priced in a great deal of it. Bloom responded that its outlook does not depend on a single project and that its modular servers can be redeployed to other sites. That is a fair point, but revenue timing still matters.

For Blue Owl, the notice lands on top of an already difficult year, with the stock down more than 40 percent over the past twelve months. As the owner and financier, Blue Owl is where the delay risk ultimately settles if Oracle's notice holds.

Could This Move the Whole Market?

One force majeure notice will not stop the AI buildout. But it does change how the market prices risk, and that can matter a great deal.

The AI infrastructure boom has been financed on an assumption: that a signed lease with a hyperscaler is almost as good as cash. Project Jupiter tests that assumption. If tenants can use force majeure to defer payment when permits fail, then lenders will look much harder at permitting risk before committing capital. That means higher spreads, more conditions precedent tied to permits, and more pressure on developers to show that the power supply is fully permitted before financing closes.

This is not an isolated case either. Industry trackers have reported that local opposition disrupted about 45 data center projects worth roughly $68 billion in the second quarter of 2026 alone, following about 75 projects worth around $130 billion in the first quarter. Combine that with the congressional attention on who pays for data center power, which I wrote about earlier this week, and a picture emerges. The constraint on AI is shifting from chips, to power, and now to permission.

Lessons for Developers

So what should developers, investors, and power providers take away from this?

First, map the entire permitting chain early, not just the power plant. That means the fuel supply, the water source, every land owner along every route, and every agency with authority over any of them, including agencies like state land offices that rarely appear on a data center checklist.

Second, treat community opposition as a schedule risk, not a public relations issue. In New Mexico, it moved from public hearings to the state Supreme Court within weeks.

Third, build redundancy into fuel and power. A single pipeline is a single point of failure, both physically and legally. Alternative routes, interim fuel options, or a phased grid connection can keep a project moving when one path is blocked.

Fourth, read the force majeure clause with permitting in mind. Whether a permit denial counts as force majeure, and who bears that risk, should be negotiated deliberately before signing, not argued about after a notice arrives.

Closing Thoughts

Project Jupiter may still be completed on time. Oracle says it is on schedule, the court has cleared the way for the air permit process, and Energy Transfer is working on the pipeline.

But the notice sent on September 24 has already taught the market something valuable. In the AI race, capital is abundant, GPUs can be bought, and even power equipment can be ordered. What cannot be bought is a permit. The companies that win the next phase of this race will be the ones that treat permitting as the first engineering problem, not the last administrative step.

Power first. And permission before power.

EnergyInnoCore Editor

Independent analysis of power, grid and policy for AI data centers. Every article lists its sources. About the editor · Editorial policy

This article is for general information only and is not engineering, legal, financial or investment advice. See the disclaimer.