Stargate's First Scare: What Oracle's Force Majeure Notice Means for the AI Buildout

The First Crack in the Foundation

Oracle sent a force majeure notice to Blue Owl Capital, the lender backing a massive data center in New Mexico. The notice shields Oracle from liability if the project fails to secure certain permits. The project is called Jupiter. It is part of Stargate, the $500 billion AI infrastructure venture announced with great fanfare at the White House.

The headline made the rounds. Oracle's stock fell about 3.5%. Financial commentators called it an attempt to wriggle out of upcoming payments. The data center, slated to provide 2.45 gigawatts of compute power for OpenAI, might not be finished by 2028.

This is the first real scare for the AI buildout. It will not be the last. But it is worth understanding what happened, why it happened, and what it tells us about the broader project of building the physical infrastructure for artificial intelligence.

The AI buildout is not just a story about chips and algorithms. It is a story about power plants and pipelines. It is a story about debt and off-balance sheet financing. It is a story about county commissions and local opposition. And it is a story about the gap between what we promise and what we can deliver.

What Happened at Project Jupiter

Project Jupiter sits on 1,400 acres in New Mexico. It is a hyperscale data center campus, one of the largest in the Stargate network. Oracle is the primary tenant. The plan was for Oracle to use the facility's 2.45 gigawatts of compute power to support its commitments to OpenAI.

The project secured $18 billion in loans. Lenders grew jittery as early as January 2026. Oracle had already laid off thousands of workers in an emergency cost-cutting spree. The company's stock had collapsed 56% from its peak.

The force majeure notice is a contractual tool. It is typically reserved for catastrophes: war, natural disaster, massive social upheaval. Oracle invoked it to shield itself from liability over the project's failure to secure development permits.

The permit problems stem from a natural gas plant. Oracle had planned to build a massive gas plant to power the facility. Developers failed to secure land use permits from state and federal authorities. Oracle cancelled the gas plant plans in May 2026.

The company pivoted to solid oxide fuel cells. It redesigned the site around up to 2.45 GW of Bloom Energy fuel cells. The largest deployment of this technology to date is an 80-megawatt project in South Korea. Project Jupiter needs 2.45 gigawatts. That is a scale never attempted before.

Oracle said "Project Jupiter remains on our planned schedule." That statement strains credulity. The company redesigned the power source, invoked force majeure, and faces ongoing local opposition. The project may still get built. But the first scare has arrived.

The Power Problem: The Grid Wasn't Built for This

The AI buildout runs on electricity. Lots of it. The median energy capacity of data centers built rose from 11 megawatts in 2016 to 130 megawatts by June 2026. Planned data centers are far larger. They consume gigawatts, not megawatts.

The U.S. power grid was not designed for this load. The average wait for new power projects in data center regions is over five years, or 64 months. Long interconnection queues leave developers with two choices: wait for grid access or bring their own power on-site.

Bringing power on-site introduces new risks. Some projects have committed to first-of-a-kind technologies: small modular nuclear reactors, long-duration battery storage, new gas plants. These approaches bypass grid queues. They also introduce delays in commercial readiness and shortages of power equipment.

The shift toward on-site generation is a direct response to grid limitations. It is also a source of execution risk. A data center that cannot secure power is a very expensive building full of nothing. Project Jupiter's pivot to fuel cells is a case study in this dynamic.

Power constraints are both an opportunity and a risk. Access to electricity determines when computing capacity can begin generating revenue. Limited availability of suitable land, local opposition, and environmental concerns complicate development further.

The International Energy Agency estimates global grid investment could exceed $540 billion in 2026. That investment is necessary. It is also slow. The AI buildout cannot wait five years for a grid connection. So it builds its own power. And that introduces risks that are difficult to model and even harder to manage.

The Financing Problem: Built on Debt and Opacity

The AI buildout is enormous. A Brookings paper projects that AI investment in data center buildings, power systems, networking infrastructure, and specialized chips will total $10.3 trillion from 2025 to 2032. That is an average of 3.63% of U.S. gross domestic product per year.

This buildout is larger relative to the economy than the major U.S. canal, railroad, electrification, highway, and telecommunications investment booms. The author of the Brookings paper, Stijn Van Nieuwerburgh of Columbia University, writes that "the projected buildout would be larger relative to the economy than the major U.S. canal, railroad, electrification, highway, and telecommunications investment booms".

The financing is migrating from transparent on-balance sheet financing by major corporations to opaque off-balance sheet financing. Joint ventures. Private credit. Securitization. Special-purpose vehicles. Lease commitments. Loan guarantees.

These off-balance sheet arrangements depend on AI companies' cash flows and collateral values. Those cash flows and collateral values are subject to uncertain AI demand, rapid technological change, timely access to power and hardware, and the continued credit quality of a small number of data center tenants.

Michael Burry, of The Big Short fame, warns that Big Tech faces $3 trillion in off-balance-sheet AI costs. He says those commitments "become real liabilities" when the boom ends. Burry has raised concern about a misalignment between the intended lifespan of a data center and the speed of AI chip development.

Goldman Sachs research found that between 33 and 37 percent of all capital expenditures by AI hyperscalers is debt finance. That does not count off-balance-sheet debt.

Van Nieuwerburgh writes that "it would be premature to conclude that AI infrastructure already poses systemic risk comparable to earlier credit booms." But he notes that "off-balance sheet structures matter … because they may make correlated exposures hard to observe before a downturn." He calls for improved measurement and transparency while the capital structure of the industry is still evolving.

Project Jupiter is a case study in this opacity. The $18 billion in loans, the force majeure notice, the redesign around fuel cells-these are visible. The off-balance sheet arrangements behind them are not.

The Community Problem: Pushback and Permits

Local opposition is a real and growing risk. At least 75 U.S. data-center projects were delayed or cancelled in the first quarter of 2026. That matches the total for all of 2025.

More than half of projects under development are in counties where earlier projects were withdrawn or where stricter requirements have been adopted. More than half are in areas that have seen successful opposition.

Around 80% of U.S. data centers under construction or planned are in jurisdictions with active or pending data-center-related legislation. Moratoriums, ordinances, zoning restrictions. These increase compliance costs. Moratoriums can halt projects that have not begun construction. Pending moratorium decisions affect about 14% of planned data-center locations.

Local elections add another source of uncertainty. Twelve percent of planned data center locations face elections that could shift the regulatory landscape.

Project Jupiter ran into "heaps of local opposition and delays." The natural gas plant was cancelled after developers failed to secure land use permits. The pivot to fuel cells is an attempt to sidestep the permitting problem. It is not clear that it will work.

The community problem is not going away. Data centers consume enormous amounts of power and water. They generate noise. They require transmission lines and pipelines. Communities are beginning to ask what they get in return. In many cases, the answer is not much beyond construction jobs and some tax revenue.

The Warning for the AI Buildout

UBS identifies nine North American projects affected by delays or cancellations. They represent roughly 7 gigawatts of initial capacity, or about 5% of the region's 130 GW announced pipeline. UBS views risks to its forecast as manageable for now. But it cautions that "resilient demand does not guarantee timely delivery".

The difference between AI demand and AI delivery matters. Demand for AI compute is strong. Delivery of that compute requires power, permits, financing, and community acceptance. The buildout can stumble on any of these. Project Jupiter stumbled on power and permits. The financing structure made the stumble more visible and more consequential.

UBS warns that "potentially slower deployment would risk pushing out revenue recognition, increase financing costs, and affect supplier order visibility." Private-market investors should prioritize brownfield projects with visible cash flows.

The first scare is a warning. It does not mean the AI buildout is doomed. It means the buildout is entering a phase where execution matters more than narrative. The easy money has been made. The hard work of building power plants, securing permits, and managing debt has begun.

What This Means Going Forward

For investors: prioritize projects with visible cash flows and proven technologies. Be skeptical of projections that assume everything goes according to plan. Pay attention to off-balance sheet financing and correlated exposures.

For developers: engage communities early. Secure power before you break ground. Be transparent about risks. The era of building first and asking questions later is ending.

For policymakers: improve measurement and transparency of off-balance sheet structures. The capital structure of the AI industry is still evolving. Now is the time to understand it.

For everyone else: pay attention to the fundamentals. The AI buildout is not a story about hype. It is a story about power, money, and people. Those who understand the physical and financial constraints will be better positioned than those who do not.

The Music Hasn't Stopped, But It's Slowing

Oracle's force majeure notice at Project Jupiter is the first real scare for the AI buildout. It will not be the last. The project may still get built. The fuel cell redesign may work. The permits may come through. But the warning is clear: the AI buildout faces power constraints, financing risks, and community opposition that cannot be wished away.

The music hasn't stopped. But it's slowing. The easy part-announcing the vision-is over. The hard part-building the infrastructure-has begun. Project Jupiter is a warning. The question is whether anyone is listening.