Everything About AI
NextFin News - The developer of power infrastructure for Oracle Corp.'s Project Jupiter data center in New Mexico has waived a Nov. 23 deadline for a state decision on a key air permit, a tactical move that buys regulators more time but does not resolve the deeper problem facing the multibillion-dollar AI campus: its 2.46-gigawatt fuel-cell microgrid still lacks a construction permit, and the natural-gas pipeline meant to feed it has slipped to a February 2027 in-service target.
Yucca Growth Infrastructure LLC is relinquishing its right to a Nov. 23 ruling from the New Mexico Environment Department on Permit No. NSR 10883, the authorization needed to build the fuel-cell microgrid that will power the Oracle and OpenAI data center campus in Doña Ana County. Rather than forcing a denial by deadline, the waiver extends the review window — a procedural step that keeps the application alive while the project's power timeline frays. The department had faced a binding statutory deadline under New Mexico's Air Quality Control Act to decide by Nov. 23; Oracle was aware of that deadline as early as May, according to a July filing by Yucca in the permit proceeding.
The permit is the linchpin of Project Jupiter's revised energy plan. In April, Yucca withdrew earlier applications for natural-gas plants and filed instead for a 2,462-megawatt microgrid built from Bloom Energy solid-oxide fuel cells — as many as 2,275 fuel-cell clusters that would supply all base-load and backup power for the campus, with Bloom operating the facility. The New Mexico Environment Department's own staff cleared a major technical hurdle in June, concluding in an air-dispersion modeling review that the microgrid would neither cause nor contribute to exceedances of applicable air-quality standards and that the permit could be issued on that basis.
But the regulatory clock has been a project concern for months. At a July scheduling hearing, a Yucca attorney told the state her client stood to lose $325 million per month beginning in November if the permit did not move, and urged regulators to accelerate the proceeding. In September, the company went further: Oracle issued a force majeure notice tied to Project Jupiter, preserving contractual protections if the campus fails to come online in 2028 as planned.
"They do not, by themselves, establish a project delay or change delivery expectations," Oracle Vice President Michael Egbert said of the force majeure notices, which he described as common on developments of this scale.
STACK Infrastructure and Blue Owl Capital, the campus developer and its owner, say they remain fully aligned and that the notice does not change their financial commitments to the project.
The market is drawing its own conclusion. Oracle shares fell 6.8% on Nov. 5 to close at $114.62, extending a decline that has left the stock roughly 54% below its June 1 high of $250.25 and trading near its 52-week low of $111.66. Bloom Energy, whose 2.45 GW of fuel-cell capacity is the centerpiece of the project's power design, dropped 17.1% on the day to $166.16, trading well below its 50-day average of $247.97 and down more than half from its 52-week high of $351.28.
There is a second regulatory cloud over the project that the waiver does not touch. The New Mexico Department of Justice has been investigating complaints that residents' names were used on supportive public comments submitted during the permit's comment period without their consent, and third parties have asked the environment department to suspend the proceeding until that investigation is complete. The proposed microgrid site sits 3.6 miles south of Santa Teresa and near the Sunland Park ozone nonattainment area, which is why the air-dispersion modeling was the department's central technical question in the first place.
The Waiver Solves the Deadline, Not the Bottleneck
The immediate question is why a developer would voluntarily give up a statutory deadline that could have forced regulators' hands. The answer is that a denial would have been worse than a delay. A denial would have sent the application back to square one, likely triggering fresh litigation and resetting a review that department staff had already found technically sound. The waiver converts a binary regulatory risk — permit or no permit — into a timeline risk. And for a project whose economics depend on a 2028 online date, timeline risk is the slower, quieter killer.
The mechanism is straightforward: data center revenue begins when power flows, not when steel rises. Oracle's July letter to the department, signed by Senior Vice President Josh Pitcock, promised roughly 4,000 construction jobs, 1,500 ongoing positions, $384 million a year in economic output for Doña Ana County during construction, and $113 million a year once operational. Every dollar of that promise is back-loaded behind the permit and the pipeline. The fuel cells cannot be commissioned without pipeline gas, and the pipeline itself is stuck in its own permitting process with an in-service target now pushed to February 2027 — a date that sits awkwardly against the campus's 2028 target and leaves little slack for a facility that needs full power at startup.
There is also a sequencing problem buried in Yucca's own filing. The application states that construction of the microgrid begins "Upon Permit Issuance," with completion expected in November 2029, yet separately lists anticipated startup around October 2026 — dates that cannot both be true. That contradiction is the clearest evidence that the project's internal schedule and its regulatory reality have diverged. A campus designed to draw about 2.45 GW cannot reach its 2028 target if its power source cannot begin construction until a permit issues and then needs years to build out 2,275 fuel-cell clusters.
The structural point is that Project Jupiter is not just late; it is exposed to a chain of dependencies that no single waiver can compress. The air permit gates the microgrid; the pipeline gates the fuel cells; the fuel cells gate the data halls; and the data halls gate the revenue that Oracle has already promised to OpenAI and booked as contracted backlog. Break any link and the 2028 date moves.
Why the Market Is Repricing Oracle's AI Power Thesis
The permit waiver lands on a stock that has already given back much of the AI euphoria of earlier in the year. Oracle's shares peaked at $250.25 on June 1 and have since fallen into the mid-$110s, a decline of more than half that has trimmed the company's market capitalization to about $346 billion. The sell-off reflects a shift in what investors are willing to pay for AI infrastructure promises: the same story that drove the stock to a record now looks exposed to the physical constraints of permitting timelines, fuel supply, and grid interconnection.
The second-order effect runs through Oracle's cloud economics. The company's AI strategy is anchored by a roughly $300 billion OpenAI commitment that helped lift its disclosed remaining performance obligations to $455 billion in the first quarter of fiscal 2026, up 359% from a year earlier. That backlog is a promise of future revenue — but revenue recognition depends on delivering capacity on schedule. If Project Jupiter's power comes online later than 2028, Oracle faces a choice between absorbing the delay, paying premium prices for interim power, or renegotiating customer commitments. Each option compresses the margin story that justified the valuation. The waiver is a small procedural event; the repricing is the market asking whether the AI data center buildout can outrun the infrastructure it depends on.
The capital intensity of that buildout is the other side of the equation. Oracle's spending on data centers has risen sharply — trailing four-quarter capital expenditures reached $55.7 billion by the end of fiscal 2026, up from $21.2 billion in fiscal 2025, according to the company's earnings release. When capex grows faster than revenue, the market starts pricing execution risk instead of growth optionality. Project Jupiter is the most visible test of whether that spending actually converts into billable capacity, and the permit waiver is the first public sign that conversion may take longer than the model assumes.
Bloom Energy is the more direct casualty. Project Jupiter is Bloom's single largest fuel-cell deployment, and the stock's 17% drop on the news reflects how tightly the company's 2026 narrative — that solid-oxide fuel cells are the standard for on-site AI power — is tied to one customer's permitting calendar. When a flagship project's timeline slips, the reference case for the entire technology class weakens, and Bloom's valuation — which had priced in rapid replication of the Oracle model across other hyperscalers — loses its cleanest comparables.
The Counter-Thesis: This Is Routine, and the Project Is Still on Schedule
The strongest argument against reading the waiver as a warning sign is that Oracle and its partners say nothing has changed. Oracle maintains that Project Jupiter remains on its planned schedule; Bloom says Oracle remains committed to its contract for 2.4 GW of natural-gas fuel-cell capacity and expects to execute on Oracle's timeline; the campus developers say they remain aligned. The environment department's staff modeling already concluded the permit could be issued. From this vantage point, the waiver is a cooperative, even constructive, step that avoids an unnecessary regulatory collision and keeps a technically sound application moving.
Force majeure notices are indeed common on projects of this scale, and Egbert is right that they preserve rights without establishing delay. The counter-thesis also has a point about the emissions math: Oracle's updated fuel-cell plan cuts nitrogen oxide emissions by 92%, carbon monoxide by 67%, volatile organic compounds by 38%, particulate matter by 83%, and carbon dioxide by 21% relative to the January design — a materially improved proposal that gives regulators a defensible basis for approval. The department has every incentive to approve a project that cuts local pollutants by those margins while bringing billions in investment and thousands of jobs to Doña Ana County.
But the counter-thesis rests on a premise the filing itself undermines: that the schedule is intact. A project whose own application lists a November 2029 completion alongside an October 2026 startup, whose pipeline has already slipped to February 2027, and whose anchor tenant has invoked force majeure to protect against a 2028 miss is a project where the schedule is the thing under stress. The waiver does not refute that; it manages it. And the pending Justice Department investigation into the integrity of the public-comment record adds a litigation overhang that no waiver resolves — if comment fraud is substantiated, the department could face pressure to reopen or redo parts of the review, adding months to a timeline that already has none to spare.
The fair read is not that Project Jupiter will fail. It is that the project's critical path has moved from construction to regulation, and regulation does not run on a developer's Gantt chart.
What Comes Next for Oracle, Bloom, and the AI Buildout
The waiver is best read as a structural signal, not a cyclical one. A cyclical delay would mean a short-term permitting hiccup that resolves and reverts; what Project Jupiter is hitting is a regime change in how AI infrastructure gets built. Data center developers can no longer assume that power will be available when the servers arrive. The bottleneck has moved upstream — into gas pipelines, air permits, water rights, and interconnection queues — and those constraints do not self-correct on a quarterly earnings cycle. The companies that priced AI growth on the assumption that capacity would materialize on schedule are now discovering that the hard part of the buildout is not buying chips; it is getting electrons to them on time.
Who is exposed and who benefits is now clearer. Oracle and OpenAI bear the delivery risk: a 2028 miss would push revenue recognition and force a reassessment of the AI capex plan that has driven Oracle's valuation. Bloom Energy bears the reference-case risk: its technology's showcase project is also its most visible point of failure. The beneficiaries are the players who can deliver power without new greenfield permits — utilities with spare capacity, developers with existing interconnection rights, and equipment vendors whose products do not depend on a single state's air-quality bureaucracy.
The forward look is specific. Watch three signals: first, whether the New Mexico Environment Department issues Permit No. NSR 10883 before year-end; second, whether the natural-gas pipeline holds its February 2027 in-service target; third, whether Oracle withdraws or narrows its force majeure notice as the 2028 date approaches. The falsifying signal for the structural-delay thesis is simple: if the permit is issued by Dec. 31, 2026, and the pipeline reaches in-service by February 2027, the delay was a one-off regulatory hiccup, not a regime shift.
Base case: the permit issues in the first half of 2027, the pipeline slips again, and Project Jupiter's power comes online well after the 2028 target, keeping pressure on Oracle's AI margin story and Bloom's replication narrative. Upside case: regulators fast-track the technically cleared application, the pipeline holds, and the campus powers up on schedule, restoring the AI infrastructure premium. Downside case: the fuel-cell design faces fresh challenges during detailed review or the comment-fraud investigation forces a partial redo of the proceeding, the pipeline permit stalls, and Oracle is forced to fall back on interim gas turbines or diesel generation — the very equipment the updated plan was designed to replace.
The permit waiver did not create Project Jupiter's problem. It simply made the problem visible.
Explore more exclusive insights at nextfin.ai.
