Oracle's two biggest AI data centers are both stuck -- not on construction, but on someone else's permit

Oracle's two largest AI data-center campuses have both run into the same kind of obstacle this fall, and it isn't the one investors usually worry about with a construction megaproject. Neither site is behind on pouring concrete. Both are waiting on a piece of energy infrastructure that belongs to somebody else -- a regulated utility in one state, a pipeline company in another -- and that Oracle cannot build or permit itself.

In Port Washington, Wisconsin, Vantage Data Centers has been building "Project Lighthouse" for Oracle since December 2025, and two of its four planned halls are already fully enclosed. The campus needs up to 1.3 gigawatts of power, and that depends entirely on American Transmission Company building a new high-voltage interconnection -- which depends, in turn, on approval from the Wisconsin Public Service Commission. ATC's first application spent ten months under review before the PSC revoked its completeness finding in August, citing "hundreds of post-filing changes" to the project's scope; ATC refiled in September, resetting the statutory review clock. Infrastructure data provider Aterio modeled three scenarios from that reset: a 180-day review puts partial power online by December 2027 and full capacity by October 2028; a 360-day review pushes that to June 2028 and April 2029.

Oracle's stock fell 1.8% the day the Aterio findings became public, and the company did not respond to requests for comment on the report -- which is itself unconfirmed by Oracle, resting entirely on the analyst firm's own modeling of a public regulatory docket.

Thirteen hundred miles southwest, Oracle's Project Jupiter in New Mexico is running into a version of the same problem with a different cause. The 1,400-acre campus is one of the flagship sites of Stargate, the AI infrastructure initiative Oracle launched with OpenAI and SoftBank early in Trump's second term, and it's designed for 2.45 gigawatts of power -- supplied not by the grid but by Bloom Energy gas fuel cells. An Energy Transfer pipeline meant to feed those fuel cells has slipped nearly six months, to a new target of February 1, 2027, after regulators repeatedly denied permits for the line. A separate air-quality permit for the fuel-cell system is still pending, with New Mexico's environment department facing a November 23 deadline to rule on it.

That last row is where Oracle's public language and Oracle's own paperwork start to diverge. A company spokesperson told reporters "Project Jupiter remains on our planned schedule... we are fully committed to New Mexico and confident in our path forward" -- while, separately, Oracle sent the site's developer, a Blue Owl Capital subsidiary, a force-majeure notice. The legal point of that notice is specifically to let Oracle defer rent payments if the facility misses its 2028 deadline for reasons outside its control; the developer still has to agree a qualifying event occurred, and Oracle remains on the hook for certain fees regardless. Filing it does not mean Oracle expects to miss the date. It does mean Oracle is formally preparing for the possibility while telling the public something more confident.

The financial backdrop raises the stakes on both delays. Bloomberg reported in late September that the $18 billion in project financing behind Project Jupiter had fallen into "stressed" territory, with the loans trading at roughly 89 to 91 cents on the dollar -- a market signal that lenders are already pricing in some risk to the 2028 date, independent of anything Oracle has said publicly. (Oracle's project-specific financing is separate from its own corporate balance sheet, which carries its own $125 billion in long-term debt and $288 billion in future data-center lease commitments -- the Jupiter loans are the site's own financing, not Oracle's general credit.) Against that, Oracle's cloud infrastructure business is growing fast enough to explain why it's building at this scale in the first place: OCI revenue jumped 121% to $7.4 billion last quarter, and Oracle holds $664 billion in remaining performance obligations still to be delivered.

Oracle's own response to the capacity squeeze, so far, has been to get more out of what it already has rather than to publicly address either site: the company recently introduced Fusion Claw, a runtime tool for governed agentic execution, and an OCI NetApp storage service aimed at hybrid AI workloads -- both pitched as making "each unit of already available compute and storage more productive" while the physical infrastructure catches up. Neither tool changes when Project Lighthouse or Project Jupiter actually comes online; both are a way to say something to customers while the permits move at their own pace.

Project Jupiter's timeline matters past Oracle's own balance sheet, because Stargate is infrastructure OpenAI is counting on for compute capacity it does not have today. A missed 2028 date on one of the program's flagship sites pushes that capacity further out for whichever of Oracle's AI customers were counting on it -- not just the line item on Oracle's own revenue schedule. Neither company has said what happens to that dependency if New Mexico's pipeline, permit, or Wisconsin's grid line slips past the dates each is currently modeling.