The force-majeure notice is not a reprieve; it is the first public confession that Oracle’s AI empire was financed before it was permitted, powered, or proved. Oracle’s Project Jupiter in Doña Ana County was supposed to turn a $300 billion OpenAI cloud commitment into 4.5 gigawatts of computing capacity, backed by roughly $18 billion in construction debt and about $3 billion from Blue Owl; now Oracle is delaying full rent under the lease while insisting the project remains “on our planned schedule.”
The cable-news version is simpler: a tech giant overreached, a force-majeure notice arrived, a partner was removed, the stock fell, and credit raters tapped the brakes. That version is for people who do not build things.
The strongest defense of Oracle deserves to be stated plainly. Hell-or-high-water leases mean that Oracle cannot terminate and must pay whether or not the power is on. That unconditional obligation is precisely what made the financing possible. Stack Infrastructure receives a guaranteed stream; roughly twenty banks could lend against it; Blue Owl could put equity behind a campus that will not be fully powered for years. The force-majeure clause gives Oracle up to three years before full rent begins. The financing package anticipated that delay. Stack earns roughly 9% annualized during the deferral and 11% once full payments begin. The deferred rent does not vanish; it extends the lease and returns with interest.
A serious operator must also respond to a community that will not accept gas turbines and diesel generators merely because a cloud contract has been signed. Oracle pivoted toward Bloom Energy fuel cells with BorderPlex Digital Assets. It absorbed the permitting fight. It kept the OpenAI commitment. The first phase remains targeted for the third quarter of 2028. The company’s spokesman is right that force-majeure notices are commonplace in projects of this size.
All of that is true.
It is also the precise reason the notice is a confession.
A financing mechanism that makes a project possible on paper does not make the project possible in the county where it must obtain permits, secure power, build a pipeline, and turn on the machines. Hell-or-high-water was supposed to make the AI build-out unkillable. Instead, it has made the rent unkillable while leaving the compute entirely conditional.
There is no off-ramp. There is only a longer on-ramp.
Oracle’s original plan to use gas turbines and diesel generators met local resistance. The company shifted to fuel cells, but the state still has not issued the necessary air permits. A proposed 17-mile natural-gas pipeline has been rejected twice because a small portion crosses state land. The workaround required another workaround, and that workaround still requires permission. Meanwhile, Oracle removed BorderPlex from the development entirely. BorderPlex had held land and power rights and had received early support from Gov. Michelle Lujan Grisham; it is now being paid to walk away from the rights it brought to the site.
That may be consolidation. It may also be what happens when a partner becomes an obstacle to the final engineering plan. But either way, it is not the clean execution of a project that has escaped risk. It is the project consuming its partners, its schedule, and its legal allowances in order to remain alive.
Oracle says the original schedule stands. The schedule just acquired three years of legal elasticity.
The market noticed. Oracle fell as much as 8% intraday before closing down about 4%. Blue Owl fell roughly 3.6%. Some New Mexico loans traded below 90 cents on the dollar, implying paper losses of at least $1.8 billion for the original lenders on that construction financing alone. Oracle’s lenders may reasonably say those prices reflect an illiquid, unseasoned asset class rather than a final judgment on the project. Construction loans can be held to maturity. A secondary-market discount is not the same thing as a default.
But illiquidity is not imaginary risk. It is what risk looks like before the balance sheet is willing to call it by name.
S&P Global Ratings downgraded Oracle in July to one notch above junk, citing its “rapidly expanding AI infrastructure business” and capital spending that has outrun operating cash flow for several quarters. Investment grade is still investment grade. It is also one notch above junk. The distinction matters to a credit analyst and does not rescue the underlying arithmetic.
The banks did not merely discover that this new asset class is hard to trade. They discovered that a single-tenant AI campus depends on a chain of permissions no lease can compel: power, air permits, pipelines, local acceptance, state land, construction, and a customer capable of absorbing the promised capacity. Earlier this year, banks struggled to sell billions in loans tied to Oracle-lease buildouts in Texas and Wisconsin. New Mexico blinked first.
Project Jupiter is not an isolated project gone sideways. It is the template.
Oracle signed a five-year, $300 billion cloud-services agreement with OpenAI without already possessing the data-center footprint required to deliver it. Then it scrambled into mega-leases in Texas, Wisconsin, Michigan, and New Mexico. The credit system treated the lease as the asset and the campus as the collateral, as if a contractual promise could substitute for electrons, permits, labor, and a community willing to live beside the machinery.
This is how an infrastructure bubble looks when the build-out meets a county.
Fifteen states are now weighing data-center moratoriums or redesigns as operators discover that gas turbines and diesel generators do not survive community review, while fuel-cell workarounds require the permits they were meant to bypass. In Little Rock and other mid-sized cities, data-center limits have made the same point in another form: a town is not an empty server rack.
The off-balance-sheet version is larger still. Filings show big tech holding roughly $3 trillion in off-balance-sheet AI commitments. Much of that exposure rests on the same hell-or-high-water logic Oracle has now exposed in public. The contract protects the landlord’s cash flow. It does not protect the tenant’s business model when the grid, the state, and the county say no at once.
You can pay and still not get the compute.
You can pay and still not get the permits.
You can pay and still not get the pipeline.
That is the bubble’s signature—not merely a falling valuation, but a cash nozzle pointed at a problem that does not take cash. The rent keeps flowing. The turbines do not spin. The permits do not issue. The pipeline does not bore through state land. The bill comes due anyway, on a longer schedule, at a higher total cost, against a credit rating that has already moved toward the edge of speculative territory.
The mistake is not that Oracle tried to build at scale. A conservative country should be capable of building large things. The mistake is treating scale as a substitute for local knowledge and calling the resulting abstraction freedom. The people who live under the proposed power system, breathe beside the generators, bear the road and water demands, and inherit the county after the financiers leave are not friction in the model. They are part of the model.
Oakeshott understood the difference between technical knowledge and practical knowledge. A spreadsheet can price a lease. It cannot know whether a pipeline route will clear state land, whether a town will accept the noise, or whether a promised industrial future is worth becoming an amenity for somebody else’s cloud. A giant corporation is its own central planner, with the same knowledge problem that the old market fundamentalists once reserved for governments.
The answer is not to replace Oracle’s concentration with a centralized state command over every server and permit. Concentrated capital and concentrated public power are the same disease in two coats. The answer is to restore the institutions that make a project answerable to the place it enters: transparent local consent, independent permitting, enforceable community benefits, public records, and power organized through member-owned utilities rather than absentee claims.
Adams-Columbia Electric Cooperative, headquartered in Friendship, serves roughly 31,560 member-owners across twelve Wisconsin counties through a board elected by those who depend on it. That is not a magic formula, and a cooperative can fail or become managerial. It is something better than magic: an institution with a place, members, and a duty that cannot be reduced to an investor’s exit.
Leave the town its life.
Project Jupiter may still reach its first phase in the third quarter of 2028. Oracle may still collect the OpenAI contract. The financing may still be refinanced, extended, or held to maturity. None of that changes what New Mexico has revealed. The most aggressive lease in the cloud can guarantee the rent, but it cannot lease its way out of a county that will not say yes.
That is the playbook the other hyperscalers should read correctly: not how to make a project too indebted to fail, but how to build one that has earned the right to exist where people must live with it.