Oracle Stargate Force Majeure, September 2026: Power Grid Delays Stall New Mexico Data Center

Oracle has invoked force majeure on its New Mexico Stargate facility and Project Jupiter, citing power-grid delays, as its FY2026 capital expenditure of $55.66 billion — 83% of revenue and the highest intensity among hyperscalers — faces questions about whether the physical grid can keep pace with a contracted AI cloud backlog reported at $664 billion.

In September 2026, Oracle invoked force majeure on not one but two large data center projects, the clearest signal yet that the physical infrastructure layer of the AI buildout is running into hard limits the contracts did not anticipate. The company cited unforeseeable circumstances — reportedly including power-supply delays and operational or environmental constraints — to halt or delay construction at its New Mexico facility, a site being developed in partnership with OpenAI as part of the Stargate initiative. A separate notice went to Blue Owl Capital Partners regarding Project Jupiter, a campus now reportedly pushed back to a 2028 completion. Multiple reports in September placed the $18 billion construction loan associated with Project Jupiter at 89 to 91 cents on the dollar, with syndication stalled — a discount that makes tangible what Oracle's force majeure language only implies.

These notices arrive against a backdrop of commitment that is historically unprecedented in Oracle's own capital history. FY2026 capital expenditure reached $55.66 billion, a 162.4% increase from the $21.21 billion spent in FY2025, and equivalent to 83% of the year's $67.36 billion in revenue — the highest capex intensity among the seven hyperscalers tracked by Slicast. The single quarter ending August 31, 2026 alone recorded $28.50 billion in property, plant and equipment payments, the largest single-quarter figure in Oracle's SEC-reported history. First-quarter FY2027 revenue came in at $19.3 billion, beating analyst estimates, and analysts reported the AI cloud backlog had reached $664 billion. The gap between demand signal and physical delivery is therefore not a demand problem; it is an execution problem.

The operational stress is visible in adjacent signals as well. Larry Ellison was reported in September to be planning a sale of up to $7.5 billion of Oracle shares, prompting questions about the cash burn associated with the AI infrastructure pivot. On September 12, however, Oracle announced he had cancelled that 10b5-1 plan entirely; no stock was sold under it, and he disclosed no further intent to sell. That announcement was filed with the SEC on September 14. In the same period, Oracle expanded its restructuring budget by $700 million, framing the outlay as part of the AI transition rather than contraction. The stock closed at $150.59 in the September 17 session, up 5.2% on the day; it now trades at $139.54, down 3.5%, though no item in coverage explicitly explains the move between those two data points.

Oracle is simultaneously pursuing multiple routes around the power bottleneck, and the breadth of those efforts is telling. It signed 1.7-gigawatt wind-energy supply agreements with RWE and Engie for Texas operations, and issued a request for proposals for a further 2 gigawatts of new renewable capacity in New Mexico — the same state where construction is currently paused. Bloom Energy supplied fuel-cell power to an Oracle data center in just 55 days, far faster than conventional grid permitting allows, illustrating why on-site distributed generation has become a strategic priority rather than a contingency. On the compute side, Oracle reported delivering 300,000 GPUs in Q1 FY2027 and is said to have renewed older GPU supply contracts at a 20% premium — evidence that AI compute demand at Oracle is pressing against available hardware inventory even as its construction pipeline encounters delays. HPE deepened networking collaboration with Oracle in September, targeting gigawatt-scale infrastructure deployments, adding another layer of vendor commitment to a buildout that is stalling at the grid, not at the supply chain.

The investment case for Oracle is genuinely two-sided. The $55.66 billion in FY2026 capex — the highest intensity in its peer group — and a contracted backlog in the hundreds of billions position the company well if it can resolve delivery bottlenecks within a reasonable window. Oracle's own commentary that the AI infrastructure buildout can be self-funded from operating cash flow provides a credible floor under the spending thesis. Against that, force majeure on two marquee projects, a construction loan trading below par, and a reported European Commission inquiry into Oracle's software-licensing practices following action against SAP represent material risks that sit below the revenue line and could surface without much warning. Three signals are worth watching closely: whether Oracle resolves the New Mexico power-supply path and resumes Stargate construction on a disclosed timeline; whether the Project Jupiter construction loan finds buyers at par or continues its discount; and whether Q2 FY2027 results show the $664 billion backlog converting into delivered capacity — a distinction between contracted promise and operational reality that the market will increasingly demand to see drawn.