Oracle OCI, October 2026: Tencent's $7 Billion AI Chip Lease Tests a Power-Constrained Buildout

Tencent reportedly agreed to lease 100,000 NVIDIA AI chips from Oracle's Southeast Asia data centers for $7 billion over five years, corroborating OCI's offshore strategy even as regulatory power delays in Wisconsin and New Mexico stall the domestic buildout behind Oracle's record $55.66 billion capital spend in fiscal year 2026.

The disclosure that Tencent has reportedly agreed to lease 100,000 NVIDIA AI chips from Oracle's Southeast Asia data centers for approximately $7 billion over five years arrived as among the most concrete proof yet that Oracle Cloud Infrastructure is becoming a preferred offshore compute corridor for Chinese hyperscalers operating under US export restrictions. The arrangement, reported across multiple outlets in October 2026, positions Oracle not merely as a domestic cloud rival to Amazon, Microsoft, and Google, but as a geographic intermediary — a node where advanced silicon that cannot legally reach mainland China can be accessed remotely, at significant commercial scale.

The strategic logic for Tencent is direct: with export controls restricting direct shipment of frontier NVIDIA chips to China, leasing capacity from a US cloud operator's Southeast Asian facilities offers a compliant path to the compute density required for large-model training and inference. For Oracle, the deal is an affirmation that its offshore build posture — less prominent in public cloud rankings but evidently stocked with high-end GPU inventory — commands a premium. At roughly $70,000 per chip over the five-year contract term implied by the reported figures, the economics suggest scarcity, not just capability, is driving price. This dynamic fits into a broader picture: Goldman Sachs projected in September 2026 that major technology companies including Oracle would collectively spend $1.2 trillion on AI infrastructure by 2027, underscoring persistent demand for capacity wherever it can be secured.

Capital expenditure for fiscal year 2026 reached $55.66 billion — a 162.4% increase from $21.21 billion in fiscal year 2025 — representing 83% of Oracle's $67.36 billion in fiscal-year revenue, the highest capex intensity among the seven hyperscalers tracked by Slicast. The largest single quarter on record closed at $28.50 billion for the period ending August 2026. These are extraordinary commitments for a company that until recently trailed its peers in cloud infrastructure scale, and they reflect the weight of Stargate — the multi-party US AI infrastructure program that has placed Oracle at the center of a national compute buildout — as well as the obligations that come with it.

Execution, however, has run into the constraint now reshaping the entire AI infrastructure cycle: power. Oracle's 1.3-gigawatt Wisconsin AI data center campus, flagged for delay risk by Aterio in September 2026, faces regulatory approval delays for grid connection that reportedly threaten customer delivery timelines targeted for 2027. More acutely, Oracle invoked force majeure on Project Jupiter, its New Mexico Stargate campus, citing regulatory and environmental obstacles to power infrastructure. The move rattled AI data center financing sentiment and prompted reports that the project's $18 billion in construction debt was trading at 89 to 91 cents on the dollar with syndication stalled. The Bloom Energy chief executive publicly dismissed force majeure concerns in October 2026, emphasizing that the 2.4-gigawatt fuel-cell contract with Oracle remained intact, and Oracle subsequently reaffirmed that commitment — but the episode illustrated how tightly construction timelines have become coupled to power-approval calendars rather than engineering schedules alone.

Oracle's response to the power constraint has been to diversify its energy sourcing rather than wait on grid capacity. The company locked in a nuclear power agreement, reported in October 2026, to supply baseload carbon-free electricity to its data center facilities — an approach IO Fund and others have described as symptomatic of a broader industry transition in which power, not compute demand, has become the primary supply chain bottleneck. On-site and near-site generation, whether fuel cells or nuclear power, is being underwritten at a scale that would have been implausible for a cloud infrastructure company three years ago. Whether these arrangements can be permitted, built, and commissioned ahead of customer commitments is the open question.

The outlook for Oracle is one of validated demand set against uncertain supply. The Tencent deal, if it closes and performs as reportedly structured, represents meaningful revenue at high utilization rates — and the offshore, export-controls-driven demand channel it opens is unlikely to contract as long as US restrictions on advanced chip exports remain in force. At the same time, the power approval delays in Wisconsin and New Mexico are not isolated setbacks; they reflect systemic grid constraints affecting the entire hyperscaler class. Three signals are worth watching: whether the Wisconsin campus secures its grid interconnection approval in time to meet the reported 2027 delivery schedule; whether Project Jupiter's construction debt is successfully syndicated — or whether the 89-to-91-cent discount widens — as a read on institutional confidence in Oracle's execution capability; and whether additional offshore leasing arrangements of comparable scale emerge, which would test how much GPU inventory Oracle has positioned outside the United States and how quickly it can be replenished.