Business
Know the Business: A 48-Year-Old Software Annuity Making a Trillion-Dollar Bet
Oracle is two companies bolted together. One is among the best businesses ever built in software: a database-and-applications franchise that sells perpetual licenses and then collects a sticky, ~22%-of-list annual support fee for decades, plus a fast-growing applications-cloud subscription book — a high-30s%-operating-margin annuity that throws off cash. The other is barely three years old: a capital-devouring utility that rents AI compute (OCI) and, to deliver it, is spending more on data centers in a single year than it earned in revenue a decade ago. Oracle organizes itself into three businesses — cloud and license; hardware; and services — and traces its lineage to operations Larry Ellison began in June 1977 [1].
The verdict. The legacy franchise is genuinely high-quality — wide-moat, recurring, cash-generative. The new OCI bet is the entire investment debate: in fiscal 2026 (ended May 31, 2026) Oracle surpassed $67 billion of revenue with about $29 billion of non-GAAP operating income, generated $32 billion of operating cash flow — and spent a net $48 billion on capital expenditures, while booking a contracted backlog (RPO) of $638 billion, up 363% in one year [2]. That backlog is roughly nine times annual revenue and is concentrated in a handful of AI counterparties — Oracle says it signed contracts with "the who's who of AI, including OpenAI, xAI, Meta, NVIDIA, AMD, and many others" [3]. Whether Oracle is cheap or expensive depends on one number that does not yet exist in the financials: the return on tens of billions of annual capex. This tab explains the engine, the moat, the capital pivot, and how to underwrite it.
FY2026 Revenue ($B)
Backlog / RPO ($B)
FY2026 Net Capex ($B)
Market Cap ($B, ~$184/sh)
Sources: FY2026 revenue, RPO and net capex — Q4 FY2026 earnings call [2]; market cap derived from staged market data (~$184.29 close, ~2.87B shares), as reported.
The single most useful framing for Oracle today: a 30%+ operating-margin software annuity is being used as collateral to fund a lower-margin, capital-hungry AI-compute business — in exchange for a backlog nine times revenue. Everything below is about whether that trade builds or destroys value.
1. The economic engine: one cash machine funding one cash furnace
Oracle's profit has always come from a simple, beautiful loop. A customer licenses the Oracle Database — "the world's most popular enterprise database," now shipping with built-in AI vector search in the 23ai release [4] — and then buys a support contract "generally priced as a percentage of the net fees paid by the customer to purchase the license." Critically, "substantially all license support customers renew their support contracts upon expiration" [5]. That is a near-100%-renewing, high-margin annuity attached to mission-critical software a customer cannot easily rip out. On top of it sits the applications cloud (Fusion ERP, NetSuite), recognized ratably over multi-year subscriptions [8].
The three reported businesses carry radically different economics. The cloud-and-license segment is the engine — software-like gross margins in the mid-70s%. Hardware is a respectable but shrinking complement, and services is a near-breakeven enabler that exists to pull through software.
Source: derived from FY2025 segment revenue and gross profit disclosures; cloud-and-license business detail in Item 7 MD and A [8]; consolidated results on p.71 [10].
The crucial shift is inside the cloud-and-license engine. Cloud services climbed from 32% of total revenue in FY2023 to 43% in FY2025 [6], and within cloud-and-license, infrastructure (OCI) overtook applications, reaching 56% of cloud services and license support revenue. OCI is a consumption meter: customers prepay a balance that is "decremented as the OCI services are consumed," like electricity [7]. The mix is migrating from a once-per-decade license sale toward metered, capital-backed compute — which is why the cost structure is changing under the revenue.
Source: revenue lines from FY2025 10-K segment disclosures and segment data, FY2021–FY2025 [8]; cloud mix percentages on p.11 [6].
Why this matters for value: the legacy support-and-applications annuity is what funds the OCI build and what an investor can underwrite with high confidence. The OCI layer is where the growth — and the capital risk — now lives.
2. The moat: deep where the data sits, thin where the GPUs sit
A moat claim is only worth making if you can name the mechanism. Oracle's defenses differ sharply by layer — and an honest read finds a genuine fortress at the database/applications layer and, at best, a being-built position in raw infrastructure.
Sources: support-renewal and pricing mechanics, FY2025 10-K p.18 [5]; database positioning p.16 [4]; "intense competition" and "low barriers to entry" p.24 [9]; multicloud growth, Q4 FY2026 call p.3 [15].
The most underrated moat development is multicloud: Oracle databases now run inside Azure, Google Cloud and AWS, and that revenue grew 404% year-over-year in Q4 FY2026 with bookings up 325% [15]. This proves the franchise is portable — even customers who picked a rival's cloud still want the Oracle database, which is the clearest evidence the database moat is real rather than a captive-hosting artifact. CEO commentary frames the strategic logic: enterprises want to run AI against "decades of rich operations data" that "already sits in an Oracle database" [15].
But be disciplined about OCI. Oracle's own 10-K concedes "intense competition in all aspects of our business" and "low barriers to entry in many of our market segments" where new competitors "frequently emerge" [9]. In infrastructure Oracle is a challenger to AWS, Azure and Google Cloud — businesses several times its size. The OCI "moat," to the extent it exists, is speed of build, access to power and GPUs, and a willingness to deploy capital aggressively — not a structural lock-in. No moat should be assumed in IaaS; it must be earned, contract by contract.
3. The capital pivot: buybacks off, data centers on, free cash flow underwater
For most of the last decade Oracle was a cash-return machine — repurchasing shares so aggressively that retained earnings went negative and equity briefly turned negative in FY2022. That era is over. Capital has been reallocated wholesale from buybacks to data centers: capex rose from ~$2.1 billion in FY2021 to $21.2 billion in FY2025 [11], then to a net cash outlay of $48 billion in FY2026 [2]; buybacks fell from ~$21 billion to $0.6 billion over the same window.
Sources: capex of $21.2B (FY2025), FY2025 10-K p.80 [11]; FY2026 net capex $48B, Q4 FY2026 call p.1 [2]; earlier capex and buyback levels per Oracle cash-flow statements, as reported.
The cash-flow consequence is stark. Operating cash flow keeps rising — to $32 billion in FY2026, up 54% [2] — but capex now outruns it. Oracle defines free cash flow on a non-GAAP, trailing-four-quarter basis [19]; on reported numbers it went roughly breakeven-negative in FY2025 (-$0.4B) and deeply negative in FY2026 (operating cash of ~$32B against ~$48B of net capex).
Sources: FY2026 operating cash flow of $32B and net capex of $48B, Q4 FY2026 call p.1 [2]; free-cash-flow definition, FY2025 10-K p.83 [19]; FY2021–FY2025 figures per Oracle cash-flow statements, as reported. FY2026 FCF is operating cash flow less net capex.
What justifies spending the entire cash flow and then some? The backlog. RPO compounded steadily from $41 billion (FY2021) to $138 billion (FY2025), then multiplied ~4.6x to $638 billion in FY2026 [2]. Management now expects only 12% of RPO to convert within twelve months and another 34% within 13–36 months [2] — versus ~33% recognizable within a year on the smaller FY2025 backlog [12]. Translation: most of this revenue lands in later years, so the capital must go in before the revenue arrives.
Sources: FY2026 RPO of $638B, Q4 FY2026 call p.1 [2]; FY2025 RPO and ~33% next-twelve-months recognition, FY2025 10-K p.107 [12]; prior-year RPO per company segment data, as reported.
To fund FY2027, Oracle guides to ~$70 billion of net capex and says it expects to raise around $40 billion in debt and equity — including a $20 billion at-the-market equity issuance [14]. This is the pivot in one line: Oracle is now an external-capital-funded infrastructure builder, not a self-funding cash compounder. The quality of the business is no longer self-evident from the cash-flow statement; it has to be underwritten on the return the backlog eventually earns.
4. How to value it: the lens is return-on-capital, not this year's P/E
Oracle screens "expensive" on trailing earnings — roughly 42x FY2025 GAAP EPS of $4.34 — but trailing P/E is the wrong lens for a company deliberately depressing near-term margins to build capacity for contracted future revenue. Three lenses are more useful, in order of importance.
Lens 1 — forward earnings against a committed ramp. Management guides FY2027 non-GAAP EPS to $8.05 (up 18% in constant currency, excluding one-time investment gains) on revenue growth of +34% in constant currency [14]. At ~$184 that is ~23x forward earnings — a market-like multiple for guided 30%+ revenue growth, if you believe the ramp. The Analyst Day framework is the bull case made explicit: a +31% revenue CAGR and +28% EPS CAGR through FY2030 [14], underpinned by an OCI revenue path management laid out as $18B → $32B → $73B → $114B → $144B over FY2026–FY2030 [13].
Source: OCI revenue trajectory provided by management, Q1 FY2026 call p.2 [13]. Forward figures are company guidance, not results.
Lens 2 — return on the capex. This is the number that decides everything and is not yet visible. Management asserts OCI carries a 30–40% margin profile and that, encouragingly, $75 billion of contracts are now "bring-your-own-hardware or prepaid" with "no degradation in margin" — i.e. customers fund the hardware themselves [15]. If true at scale, that is capital-light growth bolted onto the franchise. The bear counter is depreciation: renting GPUs that obsolesce on a 3–5 year cycle is a margin headwind, and Oracle confirmed gross margin "stepped down around 5 points" in FY2026 with a further step-down guided for FY2027 [2] [14]. Underwrite this as: does cumulative cash-on-cash return on $48B → $70B+ annual capex clear Oracle's rising cost of capital before the GPUs depreciate?
Lens 3 — relative to peers. Oracle's profitability sits mid-pack-to-strong among genuine peers (the set Oracle itself names in its competition section). It is far more profitable than the pure-SaaS players and Amazon's blended margin, but below Microsoft, the group's profit champion — and it trades at a premium multiple reflecting its growth re-rating.
Sources: peer revenue and net income per company income statements — Microsoft [20]; Amazon [21]; IBM [22]; Salesforce [23]; SAP, reported in EUR [24]; Workday [25]; Oracle FY2025 revenue, 10-K p.71 [10]. Market caps and trailing P/E derived from staged market data and latest reported GAAP net income; SAP in euros, not converted.
Source: same peer income statements as the table above [20] [21] [22]; Oracle valuation derived from staged market data and reported earnings, as reported.
Putting it together. Oracle is not obviously cheap and not obviously expensive — it is a bet, priced as one. On guided FY2027 numbers the multiple is reasonable; the question is execution and capital return, not the entry multiple. The right valuation discipline is to anchor on FCF and cash-on-cash return on capex once the data centers fill — not on a P/E built on this year's transitional earnings.
5. The risks that actually move this stock
Sources: counterparty list, Q1 FY2026 call p.1 [3]; FY2027 capex and ~$40B debt-and-equity raise, Q4 FY2026 call p.5 [14]; gross-margin step-down, Q4 FY2026 call p.1 [2]; energy-cost risk, FY2025 10-K p.31 [16]; privacy p.43 [17]; export controls p.47 [18].
The defining structural risk is the marriage of concentration and funding: a backlog leaning on a few AI counterparties, financed with tens of billions of fresh debt and equity. If even one or two large AI contracts are delayed, renegotiated, or the counterparty's own funding wobbles, Oracle is left with depreciating capacity, new interest expense, and a higher share count — the opposite of the self-funding annuity it used to be.
Two further governance/structural notes for an investor's file. First, leadership turned over during this transition: the FY2026 calls moved from CEO Safra Catz and Chairman/CTO Larry Ellison (Q1) to a new CFO (Hilary Maxson) and operating CEOs by Q4 [2]. Second, Oracle's negative retained earnings and thin book equity are an artifact of a decade of buybacks, not distress — but they do mean conventional book-value and ROE screens are close to meaningless for this name; underwrite cash flows, not the balance-sheet ratios.
6. The Oracle-specific watchlist
The industry tab lists the sector-wide signals. For this stock, these are the few numbers that would change the thesis fastest.
Sources: RPO conversion and gross-margin guidance, Q4 FY2026 call p.1 and p.5 [2] [14]; multicloud and BYOH/prepaid detail p.3 [15].
The one-paragraph thesis. Oracle is a wide-moat software annuity — sticky database, near-100%-renewing support, an entrenched applications suite — that has chosen to spend its entire cash flow, plus tens of billions of borrowed and issued capital, building AI-compute capacity against a $638 billion contracted backlog. The legacy engine is high-quality and underwritable; the OCI bet is a leveraged option on AI demand whose payoff hinges on a return-on-capital number the financials do not yet show. Buy the stock if you believe the backlog is real, the counterparties pay, and the capex earns above its cost of capital before the GPUs depreciate; avoid it if you think any link in that chain breaks. The entry multiple is reasonable on guided numbers — but you are underwriting execution and capital discipline, not a multiple.