Oracle AI Infrastructure, October 2026: Power Grid Delays, Force Majeure, and a Nuclear Hedge

Oracle's Project Jupiter force majeure and Wisconsin grid delays confirm IO Fund's thesis that power — not customer demand — is the primary gating factor for a company that spent $55.66 billion on capital expenditure in FY2026, the highest intensity among hyperscalers.

The October 2026 reports confirming that Oracle has locked in a nuclear power agreement for its data center campuses arrived against a backdrop that IO Fund had already framed with precision: power, not demand, has become the binding constraint on AI buildout. IO Fund's analysis, flagged in October 2026, noted that simultaneous delays at Oracle, Bloom Energy's data-center operations, and Applied Digital signal a structural infrastructure bottleneck that no volume of signed customer contracts can bypass. The observation reframes Oracle's recent force majeure episodes — treated by some commentary as isolated execution stumbles — as symptomatic of a grid-capacity squeeze that is industry-wide and unlikely to resolve quickly.

The scale of Oracle's commitment makes the execution risk consequential. In fiscal year 2026, Oracle spent $55.66 billion on capital expenditure, a 162.4% increase from $21.21 billion in FY2025, equivalent to 83% of its $67.36 billion in revenue — the highest capex intensity among the seven hyperscalers tracked in Slicast's XBRL-derived dataset. The largest single quarter in that record ended August 2026 at $28.50 billion. By absolute spend Oracle ranks fifth among six peers, but its intensity ratio reflects a company committing a larger share of its operating base to this cycle than any rival. The macro context reinforces the stakes: Goldman Sachs projected in September 2026 that Big Tech would collectively deploy $1.2 trillion in AI infrastructure by 2027, roughly 50% above 2026 levels.

The friction is documented at two sites. In New Mexico, Oracle issued a force majeure notice to Blue Owl Capital, its financing counterparty for the Stargate-associated Project Jupiter campus, citing regulatory and environmental obstacles affecting power infrastructure work. The project's $18 billion construction loan was subsequently reported trading at 89 to 91 cents on the dollar, with syndication described as stalled. In Wisconsin, a planned campus that Aterio described as a 1.3-gigawatt facility faces delay because grid interconnection approval has not been granted; Aterio warned in September 2026 that the schedule carries material risk, and later reporting indicated that 2027 customer delivery commitments could be at risk. Oracle shares fell 3.5% in September 2026, with coverage attributing the decline to data center delay concerns.

Oracle's response on the energy side signals a deliberate move away from grid dependence. Multiple reports in October 2026 confirmed a nuclear power agreement designed to deliver carbon-free baseload electricity to data center facilities — a direct attempt to sidestep conventional interconnection queues. The episode with Bloom Energy has also been resolved: Oracle reaffirmed a 2.4-gigawatt fuel cell contract after its force majeure notice on the New Mexico project, and Bloom Energy's chief executive publicly dismissed the alarm, describing the units as fungible within Oracle's power arrangements. The pattern — simultaneously locking in nuclear, fuel cells, and grid supply — is consistent with IO Fund's diagnosis that watts, not workloads, have become the scarce input.

The demand-side picture offers a partial counterweight. Reports in October 2026 said Tencent has signed an agreement to lease 100,000 AI chips from Oracle for a figure described as $7 billion, with the arrangement reportedly structured through Southeast Asia to remain within the bounds of US export controls restricting direct chip shipment to mainland China. The terms have not been independently confirmed, and the Southeast Asia routing detail rests on reports using hedged language. But the broad contours — a major Chinese technology company accessing GPU compute via an OCI arrangement — align with Oracle's international strategy and underscore that demand for OCI capacity appears materially unconstrained by anything other than available supply.

The picture that emerges is of a company pressing at the outer edge of its own execution capacity: spending at a ratio no rival matches, with a customer roster that includes the largest AI developers, yet dependent on permitting and interconnection timelines it does not fully control. The opportunities are specific: Oracle's nuclear positioning, if supply comes online ahead of rivals still waiting in standard grid queues, could deliver a structural cost and reliability advantage. The risks are equally concrete: the Project Jupiter loan syndication, the Wisconsin grid approval, and the broader market concern about off-balance-sheet AI financing — a structure now common to Microsoft, Meta, Oracle, and Amazon, per October 2026 reporting — remain live variables. Three signals will clarify the outlook in coming months: whether Wisconsin receives its grid interconnection approval before year-end; whether the Project Jupiter construction debt stabilises at or above par; and whether the reported Tencent deal is confirmed at its stated terms.