Here's How Oracle's "Force Majeure" Changes the AI Data Center Thesis

Oracle issued a force majeure notice tied to its $165 billion, 2.45GW New Mexico data-center project, seeking to pause rent payments to Blue Owl Capital if delays extend beyond 2028, citing regulatory risks to revenue recognition. The project is still targeted for 2028 completion, but permitting, power availability and local opposition are raising the value of operators with approved or already-built AI infrastructure. Nebius reports short-term contracts above $40 million per MW versus roughly $12 million per MW at the start of the year, while 45 projects representing $68 billion were blocked or delayed by public opposition from April through June.

Analysis

The relevant scarcity is not land or server demand but deliverable, contracted power. Operators with energized capacity can monetize AI urgency through shorter contract cycles and higher reservation economics, while late-stage projects face a nonlinear cost of delay: customers shift workloads to incumbent colo/neocloud capacity rather than wait, raising both utilization and renewal leverage for NBIS and established wholesale providers such as EQIX and DLR. The strongest benefit should accrue over 6-18 months, once constrained supply converts into disclosed backlog, revenue recognition, and higher contracted-power valuation multiples.

The market should discount promotional per-megawatt claims until operators disclose contract duration, customer credit quality, power pass-through terms, capex per MW, and cash deposits. For CIFR, IREN, and WULF, AI hosting optionality can be valuable but is not equivalent to contracted recurring infrastructure revenue; their equity remains materially exposed to bitcoin economics, financing dilution, equipment delivery, and grid-interconnection execution. A power-constrained market can paradoxically worsen their returns if incremental MW requires expensive debt/equity or forces curtailment under unfavorable utility terms.

ORCL's risk is less a demand issue than timing mismatch between customer commitments, construction milestones, and capitalized infrastructure spend. Any slippage that delays cloud capacity availability can defer OCI revenue while fixed financing and partner obligations continue, creating downside to near-term margin expectations; OWL has the inverse exposure if contractual protections remain enforceable. Consensus may be over-extrapolating scarcity into a permanent pricing regime: policy intervention, utility-led generation additions, or hyperscalers preemptively contracting gas/nuclear supply could ease the bottleneck within 12-24 months, well before all announced projects are operational.

Key Decisions for Investors

• Prefer a 6-12 month long NBIS versus a basket short of higher-beta AI-hosting/mining proxies CIFR and WULF, sized dollar-neutral. NBIS offers cleaner exposure to leased AI capacity; the short leg hedges a broad AI-infrastructure rerating while retaining exposure to execution/financing divergence. Exit if NBIS fails to show rising contracted capacity or if CIFR/WULF disclose investment-grade, long-duration AI contracts with funded capex.
• Maintain ORCL as a watch-list short or put-spread candidate into the next guidance event rather than initiating on this headline alone. Trigger only if management lowers OCI capacity/revenue timing, raises capex materially without matching backlog conversion, or identifies permitting-related milestone changes; defined-risk puts are preferable because enterprise AI demand could overwhelm timing concerns.
• For OWL, do not underwrite a benefit until the relevant lease, guarantee, and force-majeure provisions are independently clarified. Monitor whether rent accruals, reserves, or legal disclosures change at the next filing; a confirmed enforceable payment stream would be supportive, while any concession would expose the downside of financing concentrated mega-project counterparties.
• Use EQIX and DLR as lower-volatility expressions of power scarcity if institutional demand broadens beyond speculative AI-hosting names. Reassess after quarterly bookings and development yields: falling pricing, lower preleasing, or escalating power-development capex would falsify the scarcity-premium thesis.