Oracle OCI Vera Rubin Expansion and Multi-Chip Platform, October 2026

Oracle and NVIDIA are scaling Vera Rubin to multi-gigawatt capacity on OCI — where FY2026 capital expenditure reached $55.66 billion, an 83% capex intensity that ranks first among hyperscalers tracked by Slicast — while power delays and off-balance-sheet financing questions frame the execution risk.

The most visible development to land this October is confirmation from NVIDIA's newsroom — amplified by Oracle's official blog — that the two companies are scaling the Vera Rubin NVL72 platform to multi-gigawatt capacity on Oracle Cloud Infrastructure while simultaneously bringing NVIDIA's Vera CPU into OCI's general compute catalog. The moves extend a partnership that forms part of Oracle's transition from legacy database vendor to hyperscale AI infrastructure provider.

The chip additions reported in October extend well beyond the NVIDIA relationship. According to MarketScreener, Oracle is bringing Intel Diamond Rapids and Cerebras AI inference to OCI; Cerebras availability is expected in 2027, per Investing.com. Separate reports describe new partnerships with SambaNova and Positron Atlas for high-performance and efficient inference workloads respectively. The most structurally significant disclosure, however, is Oracle becoming the first hyperscaler outside Google itself to offer Google TPUs via OCI. The demand picture is further reinforced by the separately announced collaboration with OpenAI to reshape enterprise computing for the AI era; one report from Stock Titan — treated here as unverified pending corroboration — claims OpenAI used more than 100,000 NVIDIA chips on Oracle's cloud to train a model referred to as GPT-6 Astra.

The financial commitment behind that expansion is extraordinary by any measure. Slicast's compilation of SEC XBRL filings places Oracle's capital expenditure at $55.66 billion in FY2026, a 162.4% year-over-year increase from $21.21 billion in FY2025, against FY2026 revenue of $67.36 billion — a capex intensity of 83%, the highest among seven hyperscalers tracked by Slicast. The single largest quarter in the dataset, ended August 31, 2026, reached $28.50 billion. To sustain that pace, Oracle has reportedly joined Broadcom and SpaceX in seeking tens of billions of dollars from Wall Street for AI chip purchases — a posture consistent with what NDTV Profit described as hyperscalers collectively outspending operating cash flows, with aggregate capex targeting above $1 trillion. A separate analysis highlighted the growing migration of AI infrastructure spending into off-balance-sheet structures, a financing pattern that credit markets have begun to examine more carefully. On the demand side, multiple outlets with broadly consistent terms report that Tencent has agreed to lease approximately 100,000 AI chips from Oracle's Southeast Asia infrastructure for roughly $7 billion over five years, a transaction that routes Chinese AI compute demand through Oracle's offshore capacity amid US export controls. Slicast notes that deal has not been independently verified.

Power supply has emerged as the most acute operational constraint. Oracle's 1.3-gigawatt Wisconsin data center campus — for which Aterio had flagged delay risk in September 2026 — has had its grid-connection approval process restarted, with the baseline scenario for full-capacity operations now shifted to October 2028. Oracle has issued what TheStreet described as a force-majeure warning linked to AI data center power scarcity, though the item did not identify the specific contracts or sites affected. Bloom Energy's CEO responded by characterizing the company's power units as interchangeable within Oracle's Project Jupiter arrangements and dismissing concerns about Oracle invoking that provision; separately, Oracle has reportedly proposed up to $55 million in annual power-bill credits in connection with Project Jupiter, reaffirming Bloom's role. The pipeline serving that campus faces its own delays — Oracle is pursuing FERC approval for a 17.8-mile natural gas line — and reports indicate the company has meanwhile resorted to trucking natural gas to data centers. Oracle has reportedly also secured a nuclear power deal to lock in baseload clean power, consistent with its 2026 sustainability data filing. An $18 billion US data center campus reportedly came close to serious flood damage this month, adding a further layer of site-risk complexity.

Oracle's shares closed at 130.29 in the October 13 trading session, down 9% on the day; no coverage item reviewed by Slicast provides an explicit reason for that session's decline. Broader October commentary has linked AI credit concerns to pressure on both Oracle and Broadcom, though those reports frame the dynamic thematically rather than as a direct causal attribution. By the close preceding publication, shares had recovered to 130.55, a gain of 3.6%. Three signals will determine whether October's buildout narrative translates into durable returns. First, whether Wisconsin's grid approval accelerates meaningfully before the October 2028 baseline — any further slippage defers a large tranche of revenue-generating capacity and tests investor patience with an 83%-intensity capex cycle. Second, whether Oracle's financing structures, including off-balance-sheet arrangements and the reportedly planned Wall Street fundraise, hold credit-market confidence at a sustained spending level that has no direct comparable among hyperscaler peers. Third, whether OCI's multi-chip platform — spanning Google TPU, Cerebras, NVIDIA Vera, Intel Diamond Rapids, and SambaNova — delivers measurable competitive win rates against AWS, Azure, and Google Cloud, or whether catalog breadth disperses engineering focus without generating the pricing power that an infrastructure bet of this scale requires.