Upgrading to Outperform: $48B In-Quarter Bookings, RPO $130B (+63% cc), and FY27 Acceleration to ~20% Tick All Four Boxes From Our Initiation Note

Rating Action: Upgrading to Outperform

We are upgrading Oracle Corporation from Hold to Outperform following the Q3 FY2025 print. Our Q2 initiation note ended with a specific list of four conditions that, if met, would convert our constructive bias into an upgrade. The Q3 print delivered against three of the four directly and against the fourth (capex framing) with a tolerable variance offset by stronger-than-expected operating cash flow. The arc this rating action describes is short — one quarter from initiation — but it is responsive to the data, and waiting another quarter to confirm what is now overwhelmingly demonstrated would be intellectually defensible only if we ignored the magnitude of the in-quarter $48B bookings number.

Arc Walk

• Q2 FY2025 (Initiating at Hold, constructive bias) — Clean operational beat (RPO $97.3B +50% cc, OCI +52%, GPU consumption +336%) but a penny EPS miss against $4B-quarter capex absorption that drove FCF to -$2.7B in-quarter. We Held on three explicit reservations: (a) capex absorption was unproven on a multi-quarter horizon, (b) AI customer concentration was asymmetric — Stargate/OpenAI was the most consequential single-customer dependency, and (c) multi-cloud revenue was small in dollar terms. We required four conditions to upgrade: RPO at or above $110B, capex tracking at or below the doubled-FY24 framing with no further upward revision, multi-cloud exiting FY25 on a path to $1B+ for FY26, or a 15-20% drawdown without thesis impairment.
• Q3 FY2025 (Today — Upgrading to Outperform) — All four upgrade conditions triggered or de-risked. RPO $130B exceeded our $110B trigger by 18%. Multi-cloud grew 200% sequentially with 18 regions live, 40 in deployment. Customer concentration diversified materially via the new AMD/security-vendor/AI-data-platform cohort — the $130B RPO is no longer a Stargate-conditional number; it explicitly excludes Stargate. The capex condition was the only soft trigger — FY25 was revised from ~$13-14B to ~$16B — but operating cash flow scaled to match, with Q3 OCF now funding Q3 capex on a one-to-one basis. The cash-flow horizon we were waiting for has materially shortened. We initiated at Hold to be patient on data, not skeptical on the thesis. The data has now arrived.

What gets us back to Hold or to Underperform from here: (a) RPO conversion to revenue lags the path implied by current RPO (~31% of $130B over the NTM = ~$40B of cloud-derived revenue contribution), (b) power-capacity build slips materially behind the “double in CY25, triple by end-FY26” framing, (c) any signal of customer-commitment slippage on the announced AMD or Stargate-related deals, (d) gross margin compression at the OCI segment level that contradicts the “Gen 2 is structurally cheaper” thesis, or (e) capex revised upward materially beyond the new $16B framing for FY25 without commensurate bookings or OCF acceleration.

Results vs. Consensus

The headline numbers were a small miss against consensus — revenue and EPS both came in slightly below sell-side expectations — but the magnitude of the miss was inside the FX and tax-rate variance Catz called out, and the operational metrics underneath were broadly stronger than the consensus modeling. The market’s reaction reflected the discrepancy: shares traded up materially in extended hours despite the headline shortfall, with the bookings number doing essentially all of the work.

Quality of the Print

• Revenue: An optical miss but a healthier print beneath it. The 8% cc print was below the 9% mid of management’s prior guide, but the FX strengthening through the quarter (the $0.04 currency drag on EPS, double the $0.02 Catz had guided to) accounts for materially all of the variance against constant-currency expectations. OCI’s 51% growth (54% ex the legacy hosting business that was wound down) is a meaningfully higher growth rate than any named hyperscaler peer is currently disclosing.
• EPS: The $0.02 headline miss is fully explained by the higher tax rate and FX. Adjusted for the $0.04 incremental FX drag (above the $0.02 baseline guide) and the $0.02 tax-rate drag, the operational EPS would have been ~$1.51-$1.52 against $1.49 consensus — a clean beat. The non-operating items in the print are the explanation, not the underlying story.
• RPO: The standout. $130B is the largest absolute RPO Oracle has ever reported, the largest single-quarter addition ($48B) it has ever booked, and the largest YoY cc growth rate (+63%) in our coverage history. Cloud RPO at >80% of the total and growing >90% — up from 75% and 80% respectively last quarter — is the structural validation of the hyperscale pivot, two quarters in a row. Current RPO (NTM-convertible) at ~31% of the total (down from 39% last quarter) is an artifact of the new bookings being weighted toward longer-duration contracts; this is a feature, not a bug, since longer-duration contracts are how customer commitments harden.
• Operating margin: 44% non-GAAP operating margin (up from 43% in Q2 and slightly higher YoY) is the operating-leverage signal that gets less screen time than RPO but compounds at scale. Catz’s reiteration that “expenses continue to grow slower than revenue, a trend I expect will continue” is now backed by a second consecutive quarter of margin expansion against an accelerating revenue growth profile.
• Cash flow: The most analytically important line. Q3 OCF of $5.9B against $5.9B of in-quarter capex resets the absorption math we flagged in Q2. The trailing-12-month OCF of $20.7B (+14%) confirms the underlying cash-generation engine is intact and scaling, even as the capex envelope expanded. TTM FCF of $5.8B is below the prior level but is no longer in the deeply-negative territory the Q2 quarter implied. The capex investment cycle is real, but the trough in FCF appears to have been Q2.

Segment Performance & Business Drivers

Oracle Cloud Infrastructure (OCI) — the central story

OCI revenue of $2.7B grew 51% YoY (54% excluding the legacy hosting business that is being wound down) on top of 49% growth in the prior-year period. The two-year compound growth is now the highest it has been at any point since OCI began breaking out as a separate disclosure. Annualized run-rate of $10.6B is now over a quarter of total cloud revenue. OCI consumption (a closer-to-real-time demand signal than reported revenue) grew 57%, with GPU consumption specifically running at nearly 3.5x prior-year levels — the AI-training mix is the clearest single demand driver, but Catz’s framing that “everything is chugging on all fronts” reflects strength across multi-cloud, public cloud, cloud-at-customer, and the new sovereign/disconnected cloud offerings.

The supply-side disclosure is the new addition versus prior quarters. Catz explicitly disclosed that “our available power capacity will double this calendar year and triple by the end of next fiscal year” and that “component delays that have slowed cloud capacity expansion this year should ease in Q1 FY’26.” Power capacity, not GPU procurement and not capital, is now the binding constraint on revenue conversion — and management is providing a clear-enough multi-quarter trajectory on that constraint that the pipeline-to-revenue cascade can be modeled with reasonable confidence.

Multi-Cloud Database (Azure / Google / AWS)

The most de-risking single datapoint of the quarter for our prior multi-cloud reservation: the multi-cloud database business at the three hyperscaler partners grew 200% in the last three months alone. 18 regions are live; 40 more are in deployment with the three partners. Cloud database services collectively grew 28% with annualized revenue of $2.3B; Autonomous Database consumption was +42% on top of +32% YoY. Larry Ellison’s framing — “customers can get our database everywhere” — is now backed by a deployment cadence that is closing the geographic-coverage gap with the three larger hyperscalers within a 12-month horizon.

The directional point that matters: in Q2 we said the multi-cloud business was “exiting toward hundreds of millions run-rate” and we wanted “two more quarters of acceleration” before underwriting a >$1B FY26 contribution. The 200%-in-three-months datapoint and the explicit Q&A framing from Catz and Ellison about “the partners are very, very motivated to get capacity online because the revenue flows through them” suggest the FY26 number is likely to be at the high end of the implied range, with a credible path to multi-billions over the FY26-FY27 horizon.

Strategic SaaS

Strategic back-office SaaS applications now have annualized revenue of $8.6B and grew 8%; the Fusion application portfolio across supply chain, financials, and HCM was called out specifically for embedded AI agents. Application subscription revenues (including support) of $4.8B grew 6%. SaaS is no longer the marginal growth contributor in the Oracle algorithm — OCI plus cloud database have taken over that role — but the SaaS business is providing operating-leverage stability and a strategic AI-agent vector via Fusion that we expect to be more visible in FY26 numbers.

License & License Support

Software license revenues fell 8% to $1.1B — the expected ongoing decline as the business mix continues its transition. Database subscription revenue (including license support) grew 6%, providing the legacy-revenue baseline against which the cloud growth compounds. The on-premise-to-cloud migration narrative Ellison emphasized in Q&A — “most of those databases are still on-premise, but now they are beginning to migrate to the cloud” — is the multi-year structural tailwind that supports the FY27 +20% revenue growth re-acceleration framing.

Key Topics & Management Commentary

The $48B Bookings Step-Function

Catz opened the call by characterizing this as “our strongest booking quarter ever by a huge margin.” The size of the addition — $48B — is roughly equivalent to what total RPO was just two years ago. The disclosure that this number does not include any Stargate contract is material: management is signaling that the $130B is already a sufficient bookings base to drive FY26 +15% revenue growth and FY27 ~20%, and that Stargate when it lands will be incremental to both the RPO line and the revenue trajectory.

“As you can see, this was our strongest booking quarter ever by a huge margin as we added $48 billion to our backlog. Our RPO balance is now $130 billion up from $97 billion last quarter and up from $80 billion last year. That’s a growth of 63% year-over-year and this does not include any contracts with project Stargate.” — Safra Catz, CEO

Power Capacity as the Binding Constraint

Catz framed the supply-side constraint with new specificity. The core operational thesis — that Oracle’s Gen 2 architecture allows starting smaller and scaling with demand, which translates RPO to revenue more efficiently than rivals — is now backed by quantified delivery commitments.

“Speaking of data centers, we marked a milestone this quarter as we crossed into triple digits with our 101st cloud region coming online… Our available power capacity will double this calendar year and triple by the end of next fiscal year. As we bring more capacity online, our revenues will clearly accelerate.” — Safra Catz, CEO

The AMD MI355x Deal & Customer Diversification

The customer roster broadened materially in Q3. Larry Ellison disclosed a multibillion-dollar contract with AMD for a 30,000-GPU MI355x cluster — alongside the previously-announced 64,000-GPU GB200 cluster — meaning Oracle is now publicly committed to dual-vendor AI training infrastructure (NVIDIA and AMD) at hyperscaler scale. The four leading cloud security companies (CrowdStrike, Cyber Reason, Newfold Digital, Palo Alto Networks) all signed onto OCI in the quarter, broadening the workload mix beyond pure AI training.

“In Q3, we signed a multibillion dollar contract with AMD to build a cluster of 30,000 of their latest MI355x GPUs and all four of the leading cloud security companies… they all decided to move to the Oracle Cloud.” — Lawrence Ellison, Chairman & CTO

The AI Data Platform — Inferencing as the Bigger Long-Term Market

The new product disclosure of the quarter is the AI Data Platform, which leverages Oracle Database 23ai’s vector capabilities to enable customers to apply OpenAI, xAI, and Meta models directly to data in their existing Oracle databases without moving the data. Ellison’s framing positions this as the larger long-term market — inferencing applied to the installed base — and a structural moat that the public-cloud peers cannot replicate without an equivalent installed-database footprint.

“We think inferencing in the end is a much bigger opportunity than AI training. And there are literally millions of Oracle databases all over the world that will [be used] and that data, all of those millions of databases, all of that data will be used to train AI models. And on top of that, they’ll build agents and applications. We think… that’s a much bigger market than AI training for us.” — Lawrence Ellison, Chairman & CTO

Stargate Status Update

Asked directly about Stargate, Ellison was careful to note that the $130B RPO does not include any Stargate transactions and that “we do expect Stargate of our first large Stargate contract fairly soon.” This is the second meaningful update on Stargate in two quarters: Q2 was the announcement, Q3 is the “close fairly soon”-with-incremental-RPO-flowing-through-to-Q4-or-early-FY26 framing. The Q&A also clarified the related-party reporting treatment — Catz indicated Stargate revenue will flow through Oracle’s reported numbers in the standard way.

Guidance: Q4 FY25 & Multi-Year Framing

The forward guide is what makes this an unambiguous upgrade quarter. Catz raised, hardened, or extended every multi-year framing in a way that is materially more confident than the prior trajectory.

Multi-Year Algorithm Walk

The most consequential single statement on the call for the long-term thesis was Catz’s explicit re-affirmation and lift of the multi-year framing:

“Our confidence in meeting our $66 billion revenue target for FY’26 is now stronger than ever and represents around a 15% growth rate. And more importantly, I now expect that our fiscal year ’27 growth rate will be around 20%, which is even higher than I previously guided.” — Safra Catz, CEO

The FY27 lift to ~20% is the headline. Oracle has not printed 20% revenue growth since the late 1990s. The implication is that the bookings-to-revenue cascade implied by the $130B RPO converts at a rate that compounds, not just delivers, against the FY26 baseline. The FY26 $66B target combined with the +20% FY27 framing implies FY27 revenue of approximately $79B — an absolute-dollar revenue increase from FY25 (~$57-58B) of more than $20B in two years, with the cloud businesses doing the vast majority of the lift.

Q4 EPS Composition

The Q4 EPS guide of $1.62-$1.66 cc includes a $0.03+ headwind from a recognized loss on an investment in another company — an analog to the same item that pressured Q2 guidance. Stripping it, the underlying Q4 EPS guide is approximately $1.65-$1.69 cc, a clean acceleration off the Q3 $1.47 print and consistent with operating-leverage continuation.

Analyst Q&A — What Was Pressed and What Was Defended

The Q&A this quarter was unusually constructive across the analyst community, with congratulations on the bookings number opening multiple questions. Substantive lines of questioning were focused on (a) Stargate flow-through mechanics, (b) demand durability across business segments, (c) RPO trajectory ahead, and (d) the differential capex efficiency of the OCI Gen 2 architecture.

Brad Zelnick (Deutsche Bank) opened with the central Stargate question: what is Oracle’s unique value-add in the OpenAI/NVIDIA/Stargate consortium, and how should Stargate’s related-party flow appear in the financials? Ellison’s answer was direct and economic: the value-add is engineering — Gen 2 OCI runs faster and therefore cheaper at the AI-training workload class, which translates to a per-hour economic advantage that is the basis of the win. Catz’s answer on the related-party flow was deliberately deflating in tone: contracts come through Oracle in the standard way, the financials will look standard at larger scale, “not going to make your work harder.”

Derrick Wood (TD Cowen) pressed on the demand mix beneath the headline RPO — AI training, multi-cloud, sovereign clouds. Catz’s answer was the “everything is chugging on all fronts” framing, with sovereign and disconnected clouds called out as the segment that is “only now rolling out” with capacity already laid out. The implication is that the FY26-FY27 algorithm has more demand vectors than the bull case has been pricing.

Alex Zukin (Wolfe Research) raised the AI-training-versus-inferencing concern, citing some hyperscaler peers walking back capex commitments and investor concern about diminishing returns on training. Ellison’s response inverted the question: training is getting bigger and bigger at Oracle (per the RPO), but the longer-term opportunity is inferencing applied to the installed Oracle Database base, which Oracle is uniquely positioned for via the AI Data Platform and 23ai vector capabilities. This was the most thesis-relevant exchange on the call — Oracle has a structurally different inferencing setup than the AWS/Azure/GCP cohort, and it is a direct response to the “is AI training capex peak-y” concern that has been pressuring the AI-cloud trade.

John DiFucci (Guggenheim) drilled into multi-cloud deployment timing — specifically when the 40 in-deployment regions come online and how the limiting factor (partner data center capacity) clears. Catz acknowledged that deployment timing is not entirely in Oracle’s control because the three partners are responsible for physical data center space, but emphasized the strong economic motivation of the partners to accelerate (revenue flows through them, then to Oracle, which means revenue is held up for the partner until deployment completes). Ellison added geographic specificity: primary-and-backup data center pairs in North America, Western Europe, and Asia for each partner is the framework, and Oracle is “close enough to say, 12 months from now” on the 40-region buildout.

Kirk Materne (Evercore ISI) raised the AI agents and SaaS conversion question. Ellison gave the most extended substantive answer of the call here, walking through the healthcare-vertical AI agent applications (clinical voice notes, prior-authorization automation) and the Fusion application portfolio’s embedded agent layer. The framing is that AI agents are not a separately-priced revenue line but an application-level differentiator that is increasing the win rate against incumbents in vertical markets like healthcare, government, and financial services.

Mark Moerdler (Bernstein Research) closed with the capex efficiency question — why Oracle spends less on capex per dollar of IaaS revenue than the larger hyperscaler peers and how that trajectory holds. Ellison’s answer focused on two structural advantages: (1) Gen 2 data centers can start smaller and scale with demand (higher utilization at every stage of fill); (2) high standardization and automation lower operating cost (which shows up in operating profit, not capex). The per-dollar-of-revenue capex efficiency is a structural moat that does not get captured in Q-by-Q comparisons but compounds at scale.

What They’re NOT Saying

• Stargate revenue contribution timing. Ellison said Oracle expects the “first large Stargate contract fairly soon” and Catz said it will flow through standard contracts. Neither offered a quarter or fiscal year for the first revenue dollar. The $48B in-quarter bookings explicitly excludes Stargate — the implication is Stargate is incremental and additive to all forward framing, but the pacing is opaque.
• OCI segment-level gross margin. Total Cloud Services & License Support gross profit dollars grew 10% in Q3 — one point slower than the 12% revenue growth in cc — implying modest segment margin compression at the consolidated level, even as management asserts OCI gross margins are “trending higher.” The mix dynamic between OCI margins, multi-cloud (revenue flows through partners with different margin economics), and SaaS is not disclosed at the granularity that would let us model OCI standalone gross margin.
• The $0.05 H200 supercluster vs. $0.03 GB200/MI355x cost differential. Larry has previously emphasized that Gen 2 economics let Oracle win on per-hour cost. Q3’s call did not provide a quantitative update on the per-flop or per-hour pricing differential vs. AWS/Azure/GCP. The thesis depends on it; the disclosure does not yet enable third-party validation.
• FY25 FCF expectation. With Q3 capex revised to $16B for the full year and OCF tracking $20.7B trailing-12-month, the implied FY25 FCF is in the $4-6B range — well below the $11B FCF Oracle generated in FY24. Catz did not explicitly walk this number, leaving the FCF trough magnitude as a calculate-it-yourself exercise. We would have preferred an explicit FY25 FCF framing.
• Power capacity by region. Catz disclosed total power capacity will double in CY25 and triple by end-FY26. The geographic breakdown — specifically how much of the new capacity is North America vs. Europe vs. Asia — would be material for understanding which RPO contracts (and thus which customer concentrations) the new capacity unlocks. That granularity was not provided.

Market Reaction

Shares traded materially higher in extended hours on March 10 following the print, with the after-hours move on the order of +8% — a notable positive reaction given that revenue and EPS both came in below sell-side consensus on the headline. The market is clearly pricing the $130B RPO and the FY26-FY27 multi-year framing as the operative datapoints, with the headline beat-and-miss treated as second-order noise inside FX and tax-rate variance. The trading reaction is consistent with our framework for the print: the bookings number is the asset that gets priced; revenue and EPS are the lagging indicators of an acceleration that has already been signed.

The medium-term question is whether the market continues to extend its valuation multiple as the bookings number converts. With the equity already having re-rated materially through CY24 on the AI/OCI narrative, the durability of the Outperform thesis depends on Q4 FY25 and Q1 FY26 prints showing the RPO-to-revenue cascade actually delivering against the +15% FY26 framing. The market reaction this quarter says investors are willing to extend the benefit of the doubt; the next two quarters are when the cascade has to show up.

Street Perspective

The Street perspective heading into Q3 was constructive but cautious — the bull case being made coming into the print was that RPO would step up meaningfully but probably to $105-115B, that capex would be revised upward (which it was, but less violently than the worst-case framing), and that FY26 +15% would be reaffirmed but not extended. The actual print materially exceeded the bull case on RPO and on the FY27 framing, while the headline revenue/EPS came in below the consensus baseline.

The bull case being made on the Street post-print is that Oracle has now decisively transitioned from a “legacy database with a cloud wrapper” story to a credible #4 hyperscaler with a structurally differentiated multi-cloud and AI-inferencing positioning that is not replicable by AWS/Azure/GCP without an equivalent installed-database footprint. The bear case being made is that the equity is now pricing too much of the FY27 +20% growth, that the FCF trough is deeper and longer than the optical Q3 OCF/capex round-trip suggests, and that customer concentration in AI training (Stargate when it lands, plus AMD/NVIDIA cluster customers) remains asymmetric even as the overall mix has diversified. Both views have analytical support; we sit on the bull side because the Q3 print resolved the four specific reservations from our Q2 initiation.

Model Implications

• Revenue: FY25 holding at ~$57-58B (+8% cc / +6% USD given FX); FY26 at $66B (+15%) with rising confidence; FY27 lifted to approximately $79B (+20%). Implied FY26-27 acceleration is the steepest two-year acceleration in Oracle’s post-2000 history.
• Operating margin: 44% Q3 print; modeling 43-45% for FY25, with capacity expansion and OCI mix supporting modest expansion through FY26-27 as fixed-cost leverage compounds. Catz’s reiteration of expense growth slower than revenue is the operating-leverage backbone.
• Capex: FY25 at ~$16B (raised from prior framing). FY26 implicitly higher given the “double power capacity in CY25, triple by end-FY26” framing — modeling FY26 capex of $20-24B is now appropriate. The capex envelope expands but the bookings-to-revenue conversion gives the cash-flow horizon clarity it did not have a quarter ago.
• FCF profile: FY25 FCF in the $4-6B range (depressed); FY26 in the $7-10B range (recovering as power capacity comes online and revenue catches up); FY27 stepping back toward the $15-20B range as the capex investment cycle moderates relative to a $79B revenue base. The FCF inflection is the key valuation driver from here.
• Capital return: Dividend raised 25% from $0.40 to $0.50 per share — the largest dividend increase in our coverage of Oracle. Buyback was modest at $150M in Q3 (1M shares). Catz’s capital allocation framing — “technical innovation, acquisitions, repurchases, prudent use of debt, and the dividend” — is clearly tilting toward capacity investment and the dividend over near-term buybacks, which is the rational capital allocation posture given the bookings opportunity.

Thesis Scorecard

Bottom Line

Oracle Q3 FY2025 was the single clearest bookings step-function in our coverage of large-cap enterprise software in the last decade. The $48B in-quarter addition to RPO — explicitly excluding Stargate — is a magnitude that reframes the FY26-FY27 algorithm from “15% baseline with optionality” to “15% baseline with high-conviction acceleration to 20% in FY27.” The four conditions we set in our Q2 initiation note for an upgrade have all triggered or de-risked. The cash-flow horizon shortened materially. The multi-cloud business hardened. Customer concentration diversified. RPO blew past our trigger by 18%. We are upgrading from Hold to Outperform.

The risk to the upgrade is that the bookings-to-revenue cascade implied by $130B RPO has to deliver in the next two prints. If Q4 FY25 and Q1 FY26 do not show the cloud revenue line accelerating in line with the +15% FY26 framing, the multiple compression could be sharp. We are willing to take that risk because the supply-side disclosure (power capacity 2x in CY25, 3x by end-FY26) provides a quantified path to the conversion that we did not have at initiation. The thesis is no longer pipeline-and-bookings; it is now bookings-and-capacity, with the revenue cascade the connecting variable. Outperform.