Oracle issues force majeure notice on New Mexico Stargate data center

Oracle filed a force majeure notice on Project Jupiter, its 2.45-gigawatt Stargate data center campus in New Mexico, developed by Blue Owl Capital. The notice gives Oracle the legal right to stop paying rent if the building isn't ready on time. Oracle's public statement says the project "remains on planned schedule." Both things cannot be fully true. You file force majeure paperwork when your lawyers see the schedule slipping, not when they're confident it holds.

The reason for the slip has nothing to do with chips or capital. The gas pipeline from Energy Transfer slid nearly six months, to February 2027, after permits were rejected more than once. The Bloom Energy fuel cell air permit is still pending. Two of the three power dependencies are unresolved, and permit chains compound: the pipeline has to close before the fuel-cell approval makes sense, and both have to clear before grid interconnect can finalize. A six-month slip at the front of that chain does not stay six months by the time it works through the rest.

For anyone who penciled in new Stargate capacity for 2027 or early 2028, the practical consequence is that announced gigawatts are not the same as capacity you can actually rent. Blackwell racks are too expensive to warehouse. They get rerouted to whatever building already has power flowing. If Project Jupiter slips, the compute earmarked for it goes somewhere else, and "somewhere else" is already scarce, because every hyperscaler is chasing the same short list of sites with power already connected.

The less-discussed risk is concentration. If this campus eventually hosts a large share of OpenAI's training and inference workload, then a regulatory hold or physical disruption at one New Mexico site stops being a capacity shortfall and becomes a service outage that runs through every product built on top of it. The force majeure notice accidentally traces a dependency chain, from OpenAI down to a pending state air permit, that no public reliability plan has mapped.

The call: Project Jupiter will not be fully operational by its 2028 target. Oracle or Blue Owl will publicly acknowledge a slip past 2028 by the time Oracle reports fiscal Q4 2028 earnings in June 2028. Confidence is medium. The permit record on this exact site is already one of repeated rejections; hitting 2028 cleanly requires every remaining approval to clear faster than the ones that already failed. That is the less likely path.

Oracle just told the developer of a flagship Stargate site that it might not pay on time if the site misses its 2028 opening. That is the plain meaning of a force majeure notice, whatever the "on schedule" press line says. For anyone building on AI compute, the point is bigger than one campus in New Mexico: the thing that decides when giant AI data centers turn on is no longer chips or money. It is gas pipelines and air-quality permits.

Here is what force majeure means for the reader: it is the clause in a contract that lets you off the hook when something outside your control goes wrong. Oracle rents space at Project Jupiter from Blue Owl Capital. By filing this notice, Oracle is reserving the right to stop paying rent if the building isn't ready. You don't file that paperwork when you're confident. You file it when your lawyers see the schedule slipping.

This is easy to undo for Oracle and hard to undo for everyone downstream. Oracle can redeploy its money and its chips. A team that penciled in Stargate capacity for a 2027 launch cannot conjure power out of thin air.

The Skeptic

Read the two facts against each other. Oracle says "Project Jupiter remains on our planned schedule," then files a legal notice that only makes sense if it isn't. Both can't be fully true. The filing is what Oracle actually believes; the quote is what it wants printed. Lawyers file these routinely when permit risk shows up, so this isn't a crisis. It is confirmation. The gas pipeline from Energy Transfer slipped nearly six months to February 2027 after permits got rejected more than once. The Bloom Energy fuel cell air permit is still pending. The 100-gigawatt Stargate vision was announced with a President standing next to it and no plan for who signs the permits. That gap is the story.

The Compute Pragmatist

A 2.45-gigawatt campus is a lot of power. For scale, that is roughly two mid-sized nuclear reactors' worth, meant to feed one AI campus. If it slips to late 2027 or beyond, the chips earmarked for it don't sit in a warehouse politely waiting. Blackwell racks are too expensive to idle. They get rerouted to whatever building already has power flowing. So announced Stargate capacity is not the same as compute you can rent next year. The hard limit here is gas: no pipeline, no fuel cells at scale, no redundant power, no hyperscale density. Every headline about total planned gigawatts should be discounted by the permit calendar, and that calendar does not move at software speed.

The Builder

If your 2027 roadmap assumes new Stargate capacity comes online, treat this notice as your signal to line up a backup. Oracle's public "on schedule" is a holding statement. The filing is the operational reality. Permit chains compound: pipeline first, then fuel-cell air permit, then grid hookup. Slip one and the rest slide behind it. Practical move this quarter: check where your reserved capacity actually lives, and price out an Azure or GCP region that isn't waiting on Energy Transfer's pipeline. Ask your provider a blunt question. Is the power for our committed capacity already flowing, or is it contingent on a build that hasn't cleared permitting? The answer changes your whole plan.

The Safety Lens

Concentration is the part nobody stress-tested. If Project Jupiter eventually hosts a big slice of OpenAI's training and serving compute, then one campus becomes a single point of failure for a lot of the AI people depend on. A regulatory hold, a pipeline problem, or a physical disruption at one site stops being a capacity shortfall and becomes a service outage that ripples across every product built on top. The force majeure notice accidentally maps a dependency chain, from OpenAI down to a pending New Mexico air permit, that no public reliability plan has accounted for. Big numbers like 2.45 gigawatts hide how thin and fragile that chain actually is.

Where they disagree

The Skeptic says one campus slipping proves nothing about the whole initiative. The Compute Pragmatist says it proves plenty: if the flagship site can't clear permits, the 100-gigawatt total is a marketing figure, and the near-term supply the reader can actually buy is smaller than the headlines imply. That is the real split. Is this a routine legal hedge on one project, or the first visible crack in the gap between announced AI compute and delivered AI compute?

The second tension: the Builder wants a backup region today, while the Compute Pragmatist notes there may not be spare powered capacity to move to, because everyone is chasing the same too-few sites with power already flowing. If the constraint is the grid, dual-sourcing only helps until the alternative fills up too.

What this hinges on

One belief: is the bottleneck in AI infrastructure now energy permitting rather than chips or capital? The evidence in this story says yes. A six-month pipeline delay cascaded into a legal notice on a flagship site. Chips were never mentioned. Money was never mentioned. Permits and pipelines were the whole problem. If that pattern holds across other Stargate sites and hyperscaler builds, then the industry's stated capacity timelines are systematically optimistic, and the reader should discount every gigawatt figure by whatever the local permit calendar allows.

Prediction: Project Jupiter's 2.45-gigawatt New Mexico campus will not be fully operational by its 2028 target, and Oracle or Blue Owl will publicly acknowledge a schedule slip past 2028 by the time Oracle reports its fiscal Q4 2028 earnings in June 2028.

Confidence: Medium. The permit chain is already six months behind and still stacking.

Why: The gas pipeline from Energy Transfer already slipped nearly six months to February 2027 after repeated permit rejections, and the Bloom Energy fuel cell air permit is still pending, so two of the three power dependencies are unresolved before construction ramp even peaks. Oracle does not file a force majeure notice on a 2028 target if its own lawyers see the date holding; the filing is the belief, the "on schedule" quote is the press line. Permit delays compound because each approval gates the next, so a six-month pipeline slip does not stay six months once fuel-cell and interconnect approvals queue behind it. The opposite outcome, hitting 2028 cleanly, would require every remaining permit to clear faster than the ones already rejected, which is the less likely path given the track record on this exact site.

Revisit by 2028-06-30: We're right if Oracle or Blue Owl states the campus will not be fully online in 2028, or the site is visibly still under construction with power not flowing. We're wrong if the full 2.45-gigawatt campus is confirmed operational on or before its 2028 target.

The revisit date is far out because the target itself is, but the direction is set now. The permit calendar decides this, and the permit calendar is already losing.