Oracle Banks on Debt-Funded AI Build‑Out as Multicloud Database Surges 531%

A rare 20% growth double, delivered at speed

Oracle’s third fiscal quarter of 2026 landed with a data point the company has not seen in more than a decade: both organic total revenue and organic non-GAAP EPS grew 20% or better in US dollars. For a business long associated with methodical, license-driven growth, the acceleration underscores how far—and how fast—its cloud and AI strategy is reshaping the income statement.

Principal financial officer Doug Caring opened by stressing not only the numbers but the cadence. Oracle closed its books and reported just 10 days after quarter-end, a feat he framed as a proof point for its own Fusion applications. In a market now obsessed with “operational AI,” the ability to compress reporting cycles has become both a selling point and a signaling device.

Two balance-sheet decisions framed the quarter: a new equity stake in the newly separated TikTok US data business and a sizeable step-up in funding for Oracle’s AI infrastructure ambitions.

TikTok US has been carved out of ByteDance into an independent entity in which Oracle now owns 15% and holds a board seat. For now, there is no change to the revenue Oracle books as TikTok’s technology vendor; instead, the investment will be accounted for via the equity method, with Oracle recognizing its share of TikTok US earnings as non-operating income on a two-month lag starting in Q4. That income stream will be additive to existing operations.

Far more consequential for investors is the capital structure. In February, Oracle announced its intent to raise up to $50 billion in a mix of debt and equity financing this calendar year, while pledging not to issue additional bonds beyond that envelope in 2026. Within days, it had raised $30 billion via an oversubscribed book of investment-grade bonds and mandatory convertibles. The at-the-market equity component remains untapped for now. Caring reiterated that management is committed to preserving Oracle’s investment-grade rating even as it leans into a capital-intensive AI cycle.

SaaS “apocalypse” meets embedded AI

Against a backdrop of investor anxiety that generative AI could erode the value of application software, Oracle’s applications co-CEO Mike Cecilia offered a pointedly different narrative: AI as an accelerant, not an existential threat.

Cloud applications revenue rose 11% in constant currency, yielding a $16.1 billion annualized run rate. Within that, growth was broad-based: Fusion ERP up 14%, Fusion SCM and HCM each up 15%, Fusion CX up 6%, and NetSuite up 11%. Industry-focused SaaS for sectors such as hospitality, construction, retail, banking, local government, and telecom grew 19%.

If there was a single metric that spoke to durability, it was deferred revenue. Cloud applications deferred revenue climbed 14% in constant currency—outpacing in-quarter applications revenue growth and giving weight to Oracle’s claim that momentum is building rather than cresting.

Cecilia’s rebuttal to the “SaaS apocalypse” thesis rests on two pillars. First, the mission-critical nature and regulatory burden of the workloads Oracle runs—core banking systems, electronic health records, merchandising platforms, government back-office systems—are not easily replicated by a “small collection of features cobbled together and bolted on in the name of AI,” as he put it. Second, Oracle is aggressively using AI inside its own engineering organization, shrinking teams while increasing output, and embedding AI directly into its products.

The company says it has already delivered more than 1,000 AI agents inside its horizontal back-office and industry applications, at no additional cost to customers. On top of that, three new AI-native CX applications—lead generation and qualification, sales orchestration and automated selling, and a website generator that was used to rebuild oracle.com—have been rolled out. Cecilia noted pointedly that Salesforce lacks these specific products, and, unlike Oracle, lacks its own underlying cloud infrastructure and ERP/industry suites.

Healthcare provided one of the clearest illustrations of Oracle’s ambitions. The group’s AI-powered electronic health record system is now live, and management claims it is already cutting administrative overhead, increasing clinician throughput, improving patient access to care, and boosting provider satisfaction. In banking, Oracle now pitches a “comprehensive AI-powered SaaS platform” that runs across commercial, retail and investment banking, payments, compliance and AML, and front- and back-office workloads—with “hundreds” of embedded agents included.

The client roster from the quarter reads like a catalogue of displacement. In healthcare, Memorial Hermann Health System chose Fusion ERP, SCM and HCM over Workday. The University of New South Wales did the same. Gray Media, Investec Bank, HID Global, Ethiopian Shipping and Logistics, and “a major Wall Street bank” selected Fusion suites over SAP, with the latter standardising on Fusion ERP across all business units and abandoning SAP “full stop.” The JM Smucker Company, Westfield Insurance and Mitsubishi UFJ Financial Group added Fusion modules; Loudoun County Public Schools and Zain KSA Kuwait expanded their Oracle footprints.

More telling than the wins may be the “go-lives.” More than 2,000 customers moved into production on Fusion and industry applications in the quarter, with median time to go-live continuing to fall. Examples ranged from Hearst (expanding ERP with EPM and HCM) to Emirates Health Services (HCM for HR, payroll and talent), Niagara Bottling (migrating SCM from on-prem ERP to Fusion) and Seadrill (ERP, HCM, SCM and EPM).

Cecilia also cast Oracle’s AI agent strategy as a way to defend and deepen its place at the center of enterprise workflows. The new AI Agent Studio inside Fusion allows customers and partners to build their own agents—not just on top of Fusion data but across Oracle industry apps and third-party systems—while benefiting from quarterly upgrade and security cycles. The bet is that “data gravity” will keep AI innovation anchored close to Oracle’s system-of-record databases and apps.

Multicloud database: from “any hardware” to “every cloud”

If applications are the visible face of Oracle’s transformation, Clay Magouyrk’s comments on infrastructure and data suggest where much of the future profit pool could lie.

Multicloud database revenue surged 531% year over year. AI infrastructure revenue climbed 243% over the same period. Both lines, he stressed, are in a world where demand still outstrips supply.

Historically, the Oracle Database could run on many hardware and OS combinations, but its cloud incarnation lived only on OCI. Over the past two years, Oracle has methodically loosened that tie, cutting multicloud deals first with Microsoft, then Google, and lastly Amazon.

This quarter marked an inflection: global region coverage across all three partners. Oracle now has 33 regions live with Microsoft, 14 with Google, and went from two AWS Oracle Database regions at the start of Q3 to eight by quarter-end. It expects to exit Q4 with 22 AWS regions live. For an installed base accustomed to running Oracle Database in other clouds, this expansion effectively unlocks what Magouyrk called an “enormous backlog of demand.”

AI is intensifying that pull. The rapid advancement of model coding abilities and “agentic” behaviours is pushing customers to move their most valuable data into Oracle’s cloud services to access vector embeddings, model hosting, and advanced security, all while keeping data close to the agents operating on it. For those still on-premise, the message is blunt: until that data moves to the cloud—often via Oracle’s multicloud database services—it cannot fully participate in the AI wave.

The financial appeal of this portfolio is not subtle. While the AI infrastructure business is targeting gross margins in the 30–40% range and reported 32% this quarter for newly delivered AI capacity, the multicloud database services carry margins Magouyrk described as “more in the 60–80% range” and are “growing very, very rapidly.” Combined, they are lifting OCI’s overall margin profile even as capital intensity spikes.

AI infrastructure: 10GW secured, 400MW delivered, 29B contracted

To understand the bet embedded in Oracle’s balance sheet, one has to follow the power.

The company and its partners have now secured more than 10GW of power and data center capacity slated to come online over the next three years. More than 90% of that capacity is already fully funded by partners, with the remaining slice expected to be locked down this month. That structure is critical to Oracle’s promise of “uncoupling” CapEx from Oracle’s own capital requirements.

Operationally, the group has been quietly re-engineering its supply chain. Over the past year, it has standardised data center designs, tripled manufacturing sites, and increased rack output fourfold. It has also scaled installation processes to allow multiple phases of delivery in parallel, cutting the time from rack delivery to revenue by a double-digit percentage in just a few months.

On the funding side, Oracle is experimenting with business models designed to preserve its own cash while still capturing AI demand. Since its last earnings call, it has signed more than $29 billion of infrastructure contracts across multiple customers using variations of “bring your own hardware” and upfront customer payments. These deals are incremental to other AI contracts signed in the quarter and, crucially, do not require Oracle to raise additional debt or equity to expand capacity.

The result: in Q3 alone, Oracle delivered more than 400 megawatts of AI capacity to customers, with 90% of that committed capacity handed over on or ahead of schedule. Capacity delivered is profitable on day one; the drag on margins, Magouyrk argued, comes from the volume of capacity under construction at any given moment. As that construction converts into live, contracted workloads, the model should increasingly resemble a high-fixed-cost, high-utilisation utility—only with software-like incremental margins.

AI infrastructure spend is neither monolithic nor limited to GPUs. Oracle’s AI data centers also underpin large volumes of general-purpose compute, high-performance and blob storage, load balancing, identity, and security services. Those adjacent services typically account for 10–20% of total AI infrastructure spend and often carry higher margins than the GPUs themselves. Layer on top the multicloud database services consumed by AI-heavy workloads, and the economics begin to look more attractive than a narrow focus on accelerator gross margins might suggest.

Sovereign AI and Alloy: drawing new boundaries

One of the more nuanced themes running through the prepared remarks was sovereignty—of data, of operations, and of governance.

Cecilia argued that last year’s notion of “sovereign cloud” as merely keeping primary data in-country has already been overtaken by events. Governments and regulated industries now expect sovereign data, sovereign operations, and sovereign contracting. In practice, that means primary and backup data in-country or in-jurisdiction, local operational control, and contracts structured to reflect local legal and political priorities.

Oracle’s answer is Alloy: a model that allows partners—whether nation-states or large enterprises—to run a full-stack instance of OCI, with the complete catalogue of OCI services, and, by extension, Oracle’s applications and AI Data Platform, inside their own data centers or controlled facilities. That stack can scale from as few as three racks to hundreds, and the “line of sovereignty” can be drawn around a country or around a cross-border industry ecosystem, such as healthcare or retail in a regional bloc.

Because Alloy delivers the entirety of OCI rather than a curated subset or “edge” zone, Oracle can layer its full application portfolio atop these sovereign clouds, potentially lifting overall margin mix. For investors, the strategic implication is that the AI infrastructure Oracle is funding today could become the substrate not only for hyperscale generative AI, but also for politically sensitive, high-value sovereign workloads where switching costs are even higher.

A halo beyond GPUs

Although the AI conversation tends to gravitate to GPUs and data centers, Oracle’s executives repeatedly returned to the “halo effect” AI investment is having on the rest of the business.

Cecilia described three main channels. First, the close physical proximity of AI models and Oracle’s applications in OCI allows the company to embed “very high-quality AI services” directly into its SaaS suites, accelerating customer time-to-value and softening one of the most common criticisms of AI projects—that benefits accrue too slowly. Second, OCI’s cost and performance profile is being used as a “budget creator,” allowing enterprises to fund application modernization and AI projects by first cutting infrastructure costs. Third, the combination of OCI, AI tooling, Fusion apps, and industry suites is shifting customer conversations from single-application deals to multi-product “ecosystem automation” projects—precisely the sort of engagements that drive higher lifetime value and stickier relationships.

Tech infrastructure wins from the quarter illustrated that halo. Lockheed Martin chose OCI high-performance compute to scale AI workloads. Activision Blizzard expanded its Oracle Database footprint via Oracle Database at Azure. Air France-KLM reported a multi-cloud win with Oracle Database delivering 13x performance at significantly lower cost. Rhombus adopted OCI for AI video and security workloads, while Lucid Motors turned to OCI core services to support its European expansion. Claro Brazil selected OCI Alloy for sovereign AI; Infomart in Japan moved a mission-critical B2B platform onto OCI.

Taken together, these threads show a company trying to move beyond the traditional cloud playbook. Oracle is not simply renting out compute and storage; it is positioning itself as the architect and operator of AI-enabled, industry-specific ecosystems—from healthcare to banking—backed by a capital structure and multicloud reach designed to turn what is now a rush of demand into a long-duration stream of recurring, and increasingly profitable, revenue.