Oracle Force Majeure Notice Exposes AI’s Power Bottleneck
Oracle issued a force majeure notice after power delays threatened Project Jupiter, its planned 2.45-gigawatt AI data center campus in New Mexico. The Oracle force majeure notice does not cancel the project or prove that its opening will slip. It does reveal a widening conflict between AI infrastructure contracts and the slower systems needed to supply their electricity.
Project Jupiter is expected to support OpenAI through the Stargate infrastructure program. STACK Infrastructure, owned by Blue Owl Capital, is developing the campus. Oracle remains the main tenant and says construction is following its planned schedule.
The notice matters because Oracle has attached enormous financial commitments to capacity that does not yet exist. OpenAI and other customers want more computing power, while Oracle needs completed buildings, energy equipment, permits, fuel, and transmission infrastructure before it can recognize much of that demand as revenue.
That creates the central reversal. AI companies can order processors and sign cloud agreements faster than developers can secure dependable power. The scarcest component in Oracle’s expansion plan is no longer necessarily the GPU. It is an operating site with enough electricity to run thousands of them continuously.
What the Oracle Force Majeure Notice Actually Changed
Oracle has reserved a contractual defense against power-related delays, but it has not abandoned Project Jupiter.
Oracle sent the notice to a Blue Owl unit developing the New Mexico campus, according to a power-delay report published by Reuters. The report said potential delays in securing power prompted the action.
Force majeure is a contract provision covering extraordinary circumstances outside a party’s reasonable control. Its legal effect depends on the contract, the triggering event, and whether the affected party followed required procedures.
A notice is therefore different from a confirmed default. It can preserve contractual rights while the parties continue working toward completion. It can also begin a dispute over who accepted a particular development risk.
Blue Owl said the notice did not alter the project’s financial commitments. The company also said the parties remained aligned on the multiyear development. Oracle separately maintained that Project Jupiter remained on its planned schedule.
Those statements narrow what can be concluded. There is no verified evidence that Oracle has canceled the campus, ended its relationship with STACK, or withdrawn its commitment to New Mexico.
The immediate dispute instead concerns protection from costs if power problems delay occupancy. Reported lease terms indicate that Oracle could seek to postpone rent commencement for three years if both sides recognize a qualifying force majeure event.
That outcome is not automatic. The developer would need to accept the claim, or the parties would need to resolve any disagreement through their contract’s dispute process.
Oracle would reportedly continue covering certain project costs during a deferral. It would also remain responsible for rent over the lease’s full term once payments started. These details make the notice more defensive than terminal.
The distinction matters for investors and cloud customers. A canceled facility would signal weakened demand or a fundamental break in the project. A payment deferral would instead shift the timing and allocation of construction risk.
Project Jupiter’s planned scale explains why that timing matters. The campus was designed for 2.45 gigawatts of capacity, according to the Bloomberg account. A site that large needs an energy system comparable to infrastructure serving a substantial metropolitan area.
It cannot rely on an ordinary commercial connection. The project needs dedicated generation, fuel delivery, electrical equipment, local approvals, and systems that can operate under demanding uptime requirements.
That makes power a construction dependency rather than a monthly utility purchase. If one critical component arrives late, completed data halls may remain unable to host revenue-producing computing equipment.
The notice turns that physical dependency into a financial question. Who pays while a largely built AI campus waits for usable energy?
AI Demand Has Moved Faster Than Energy Infrastructure
Project Jupiter shows that contracted AI demand and deliverable computing capacity are now moving on different clocks.
Oracle’s cloud backlog suggests that customer interest remains strong. Its August 2026 quarterly filing reported $664 billion in remaining performance obligations.
Remaining performance obligations represent contracted revenue that has not yet been recognized. They are not the same as immediately collectible sales, guaranteed profit, or cash in Oracle’s bank account.
Oracle expected to recognize about 13% of that backlog during the following 12 months. Another 37% was expected during months 13 through 36, with 34% scheduled during months 37 through 60.
That timetable makes infrastructure execution central to the company’s financial story. Oracle cannot convert a large cloud contract into revenue on schedule unless it installs equipment and makes the promised capacity available.
The same filing disclosed $288 billion in additional lease commitments. Most related to data center arrangements expected to begin between fiscal 2027 and fiscal 2029, with terms lasting 15 to 19 years.
Those leases were not yet reflected on Oracle’s balance sheet because they had not commenced. Their scale still shows how much future capacity Oracle has contracted before the underlying sites become operational.
Oracle also reported $34.2 billion in unconditional purchase and other obligations lasting longer than one year. These commitments primarily covered cloud infrastructure components and data center power arrangements.
The company’s commercial opportunity is therefore matched by a long chain of physical obligations. It needs processors, networking equipment, buildings, cooling systems, electricity, financing, and customers to become available in the correct sequence.
Power is particularly difficult to accelerate. Chip manufacturers can expand production, although doing so takes time. Electricity projects also face utility planning, land access, environmental review, equipment queues, and public participation.
The national figures show why those queues are becoming crowded. A federal estimate placed United States data center consumption at 176 terawatt-hours during 2023.
The Department of Energy projected consumption between 325 and 580 terawatt-hours by 2028. At the upper end, demand would more than triple within five years.
That growth does not spread evenly across the country. A single hyperscale campus can create concentrated demand in an area where the existing system was never designed for it.
Utilities must determine whether new generation and transmission can serve the load without weakening reliability. Regulators must decide who finances those upgrades and who bears the risk if projected demand disappears.
Local communities also want evidence that projects will not raise household bills, consume scarce water, or create pollution without delivering promised benefits. Those concerns can influence permitting schedules even when national policymakers support AI construction.
This creates pressure on every participant. Developers need permits before lenders release capital. Utilities need credible load forecasts before building infrastructure. Technology companies need firm completion dates before promising capacity to customers.
Oracle sits in the middle of those dependencies. Its customer agreements create urgency, but its infrastructure partners cannot eliminate regulatory timelines through a cloud contract.
Project Jupiter’s Power Plan Keeps Changing
The project’s search for dependable electricity has already forced a major technical redesign.
Project Jupiter originally proposed natural gas turbines and diesel backup generators. Oracle later replaced that design with a fuel-cell microgrid using technology from Bloom Energy.
A microgrid is a local electricity system capable of supplying a defined site. In this case, fuel cells would convert natural gas into electricity through an electrochemical process rather than conventional combustion.
Oracle’s updated power plan said the new design would reduce emissions and water use compared with the previous proposal. That is Oracle’s assessment, not an independent finding about the completed campus.
The change addressed some objections surrounding the earlier turbine plan. It did not remove the project’s dependence on fuel infrastructure, regulatory approval, or timely equipment delivery.
Fuel cells can provide on-site generation without waiting for a conventional grid connection sized for the entire campus. That makes them attractive to AI developers facing long utility queues.
However, fuel cells still need a reliable natural gas supply. Their installation also requires electrical integration, air-quality review, safety controls, and enough equipment to support an unusually large load.
The proposed gas pipeline serving Project Jupiter encountered its own problems. New Mexico’s state land commissioner rejected permits affecting a planned route, adding uncertainty to the fuel-delivery schedule.
The pipeline developer subsequently faced a delay in its expected service date. That setback is important because the campus cannot solve a grid constraint with on-site generation if fuel cannot reach the generators.
The sequence illustrates a recurring infrastructure problem. Replacing one bottleneck often creates a different dependency rather than removing the original risk.
A grid connection depends on utilities, transmission capacity, and regulatory cost allocation. A dedicated generation system depends on fuel delivery, equipment production, air permits, and site approvals.
The fuel-cell approach still offers potential advantages. Modular units can be installed in phases, which could let Oracle energize sections of the campus as buildings become ready.
Phased installation also aligns better with a gradual deployment of servers. Oracle would not necessarily need the full 2.45 gigawatts on the first operational day.
Yet phased construction introduces scheduling complexity. Each usable data hall needs adequate generation, cooling, networking, backup systems, and safety approvals before customers can deploy production workloads.
A partial opening would therefore help only if the available energy matches the computing capacity promised under Oracle’s customer contracts. Public disclosures do not provide that detailed alignment.
Oracle’s insistence that the project remains on schedule may reflect confidence in alternative routes or revised construction phases. The force majeure notice suggests the company is also preparing for a less favorable outcome.
Those positions are not necessarily contradictory. A business can expect timely completion while preserving its contractual rights if an external dependency fails.
The problem is informational. Investors cannot independently confirm the construction schedule, the pipeline’s critical path, or the conditions needed for Oracle’s rent obligations to begin.
That uncertainty makes Project Jupiter a test of execution rather than demand. Oracle has customers seeking capacity. The open question is whether its energy plan can turn buildings and processors into usable cloud infrastructure on time.
The Conflict Is Between AI Contracts and Physical Reality
Oracle’s expansion model assumes that long-term financial commitments can be synchronized with infrastructure exposed to local delays.
The AI cloud market rewards companies that secure capacity early. Waiting for a site to receive every permit before signing customers could leave Oracle behind larger rivals.
Amazon, Microsoft, and Google can distribute workloads across extensive cloud regions and existing energy relationships. Oracle has a smaller infrastructure footprint, even as it pursues some of the industry’s largest AI agreements.
Oracle’s strategy therefore relies heavily on future campuses. That produces rapid reported backlog growth, but it also concentrates execution risk in projects that require years of construction.
Project Jupiter brings that tension into view. Oracle’s customer demand sits downstream from OpenAI and other model developers. Its lease obligations sit upstream with developers, lenders, utilities, and equipment suppliers.
If the campus opens on schedule, the structure can work well. Oracle receives computing revenue while infrastructure partners collect contracted payments over long lease terms.
If power arrives late, revenue and occupancy costs can separate. Oracle could face payments before it has usable capacity, while developers and lenders still expect returns on committed capital.
The Oracle force majeure notice appears designed to limit that mismatch. It gives the company a basis to argue that extraordinary power delays should postpone some payments.
Developers see the risk differently. They finance construction partly because a large technology tenant has accepted a long commitment. If that tenant can delay payments, capital providers bear more schedule risk.
That is why Blue Owl emphasized that the project’s financial commitments had not changed. The statement reassures lenders and investors without resolving how the force majeure provision would operate.
The argument is bigger than one lease. AI infrastructure financing increasingly uses special-purpose projects supported by long-term contracts from technology companies.
Those contracts help developers raise money without placing every construction cost on a cloud provider’s balance sheet. They also create dependencies among customers, tenants, developers, private-credit funds, utilities, and energy suppliers.
A delay can move through that chain. The tenant postpones occupancy, the developer waits for rent, and lenders reassess the value of project debt.
The market reaction reflected that concern. Oracle and Blue Owl shares both declined after reports of the notice, although a single trading session cannot establish long-term project economics.
The skeptical view is that Oracle is treating a foreseeable power challenge as an extraordinary event. Large data centers obviously need energy, and New Mexico approvals were always part of the development process.
From that perspective, power availability belongs in ordinary construction risk. Calling it force majeure would weaken the certainty that makes long-term project financing possible.
The opposing view is that specific permit denials or government actions can qualify even when the general category of risk was known. The legal result depends on contract language that is not public.
Neither interpretation has been conclusively established. The notice is not proof that Oracle’s claim will succeed, and the company’s schedule statement is not proof that every dependency has been resolved.
The safest conclusion is narrower. Oracle has identified enough uncertainty to preserve a contractual remedy, while all parties continue presenting Project Jupiter as an active project.
Power Scarcity Is Becoming a Cloud Competition
Cloud providers increasingly compete for deliverable megawatts, not just customers, processors, or software features.
AI models require dense clusters of accelerators connected through high-speed networks. Training jobs and large inference services can keep that equipment running at high utilization for extended periods.
That operating pattern creates a large, steady electrical load. Cooling systems, storage, network equipment, and backup infrastructure add to the servers’ direct consumption.
New facilities also want electricity sooner than conventional generation and transmission projects can provide it. Grid construction often requires more time than fitting out a data hall.
The result is an emerging market for energy-secured sites. A cloud provider with available processors but no energized building cannot sell the computing capacity customers need.
This changes the meaning of scale. Previous cloud expansion often involved adding servers within an existing region. Current AI projects can require entirely new power systems measured in gigawatts.
Amazon, Microsoft, Google, Meta, and Oracle have pursued nuclear contracts, renewable projects, grid agreements, and on-site generation. Each approach balances speed, reliability, cost, emissions, and regulatory exposure.
Grid power can offer efficient system-wide generation, but connection queues can be long. Gas generation can arrive closer to the load, yet it faces pipeline and environmental constraints.
Nuclear plants provide dependable low-carbon electricity but require long development timelines. Renewable generation can be built relatively quickly, although large AI clusters also need storage or firm backup capacity.
Fuel cells occupy another position in this mix. Their modular design supports on-site deployment, but large installations still depend on natural gas and extensive supporting infrastructure.
Project Jupiter therefore represents a broader industry experiment. Oracle is attempting to assemble computing and energy capacity at a scale that local systems were not built to supply quickly.
The company’s competitors face similar pressure, but their starting positions differ. Larger cloud operators have more regions, greater cash generation, and wider infrastructure portfolios.
Oracle’s large backlog gives it a strong incentive to expand rapidly. Its lease commitments also reduce room for prolonged delays across multiple projects.
This is why the warning reaches beyond Oracle shareholders. OpenAI depends on partners to provide the infrastructure behind its services, while enterprises depend on those services for production AI workloads.
A delayed campus does not immediately interrupt an existing application. It can affect when additional model training, inference capacity, and enterprise cloud resources become available.
Capacity constraints can influence allocation and deployment decisions before customers see an outage. Providers might prioritize large contracts, delay new regions, or shift workloads to more expensive locations.
Enterprise buyers should therefore evaluate more than announced GPU counts. The relevant question is whether a provider can deliver sustained, powered capacity in the regions and time frames a deployment requires.
For developers, the issue appears indirectly through quotas, waiting lists, capacity reservations, and regional availability. Those operational signals reveal more than a distant campus rendering.
Knowledge workers will not track pipeline permits during normal AI use. They may still encounter the consequences through slower product expansion, restricted features, or higher organizational costs.
The Oracle force majeure notice makes that chain visible. Energy planning has become part of cloud product reliability, even when the end user never sees the power source.
Three Signals Will Show Whether Oracle’s Warning Spreads
The next evidence should come from the project schedule, Oracle’s financial disclosures, and the financing terms used across new AI campuses.
The first signal is Project Jupiter’s fuel and power schedule. Oracle needs clear progress on the gas pipeline, fuel-cell installation, and approvals required for phased operation.
A firm pipeline service date would strengthen Oracle’s claim that the campus remains on schedule. Additional permit setbacks would make its contractual warning look more consequential.
The second signal is the timing of Oracle’s leases and remaining performance obligations. Its filings should show whether data center commitments begin when expected and whether backlog converts into revenue.
Investors should watch changes in not-yet-commenced leases, power-purchase obligations, capital spending, and projected revenue recognition. None of these measures alone will confirm Project Jupiter’s status.
Together, they can show whether infrastructure is catching up with contracted demand. A rising backlog alongside repeated capacity delays would widen the gap between commercial promises and physical delivery.
The third signal is how developers and lenders price schedule risk in later projects. New contracts may assign power responsibility more explicitly or require larger guarantees from technology tenants.
That response would indicate that Project Jupiter changed financing practice even if the campus ultimately opens on time. It would also raise the cost of shifting infrastructure outside a cloud provider’s balance sheet.
Conversely, unchanged financing terms and continued lender demand would suggest that the market views the notice as a project-specific precaution.
Readers should resist treating every force majeure notice as evidence of an AI crash. Project delays can occur during a durable expansion cycle, especially when construction reaches unfamiliar scale.
They should also avoid dismissing the notice as legal housekeeping. Oracle has hundreds of billions in backlog and future lease commitments tied to infrastructure that must receive dependable energy.
The essential test is conversion. Oracle must transform contracts into energized capacity, then transform that capacity into recognized cloud revenue.
Project Jupiter remains under development, and Oracle says its schedule is intact. Blue Owl says the financial commitments remain in place.
What remains unresolved is whether the power system will arrive when the computing contracts require it. The Oracle force majeure notice is the clearest sign that this timing can no longer be assumed.
Over the coming months, watch the pipeline schedule, fuel-cell approvals, and Oracle’s lease disclosures. Those signals will show whether Project Jupiter is overcoming a local delay or exposing a repeatable weakness in AI construction. For enterprise buyers, the practical response is to question delivery dates, regional capacity, and contingency plans before depending on promised infrastructure. For investors, backlog should be evaluated beside power access and lease commencement, not in isolation. The AI buildout still has customers, capital, and processors. Its next constraint is whether energy projects and public approvals can match the speed of those commitments.
