Business

Oracle Is Now Two Businesses Wearing One Ticker

For 40 years Oracle was one of the finest businesses in technology: a near-monopoly database, an applications suite that companies could not rip out, and ~90% gross-margin license support that compounded into a cash machine so reliable management spent a decade shrinking the share count by half. That business still exists, and it is still wonderful.

Bolted onto it now is a second, opposite business — an AI-infrastructure landlord that rents GPU capacity, devours capital, runs at utility-like margins, and is funded with debt and freshly issued equity. The two halves have inverted economics. Understanding Oracle means refusing to let the consolidated income statement blend them, because the market has stopped blending them — it is paying a hyperscaler price (~9× EV/sales) while consolidated returns on capital are falling.

Source: Oracle FY2026 reported results (June 10, 2026); ROIC and EV/Revenue derived from reported financials at the current ~$184 share price. Cloud crossed half of revenue in FY2026 per management.

1. The two engines, side by side

The single most useful thing an investor can do with Oracle is stop reading one P&L and start reading two. Below is the same company split by economic character. The numbers are directional — Oracle reports revenue by offering but not full segment margins — yet the contrast is the point.

Source: Revenue split estimated from FY2025 offering disclosure (cloud services & license support $44.0B + license $5.2B for the annuity; OCI is the fast-growing slice inside cloud). Margin and ROIC characterizations from management commentary and reported financials.

The annuity is a toll road: write the software once, collect support and subscription fees forever, reinvest almost nothing. The infrastructure business is a property developer: buy land (data centers), buy inventory (GPUs that depreciate fast), borrow to do it, and earn a spread on rented capacity. One throws off cash; the other consumes it. They share a brand, a sales force, and — crucially — a database that is the bridge between them.

2. Where the money is actually made

Oracle's profit has always come from recurring, high-margin software, not from selling boxes or hours. The teal slice below — cloud services and license support — is the economic heart: it is subscription-like, it renews, and the license support portion within it carries gross margins near 90% because the R&D was paid for years ago.

Source: Oracle FY2025 10-K, consolidated statement of operations. FY2026 offering detail not yet itemized; total FY2026 revenue $67.4B with cloud above half.

But notice what AI infrastructure is doing to the blended margin. Gross margin has fallen from 72.8% (FY2023) to 65.8% (FY2026) — a ~5-point step-down in FY2026 alone — and management guides it lower again in FY2027. This is not deterioration of the software; it is mix. Renting GPU capacity means depreciating tens of billions of hardware and paying for power, so every incremental infrastructure dollar arrives at a much lower gross margin than a software dollar.

Source: Derived from Oracle reported income statements, FY2023–FY2026.

The investor's central economic question lives in this chart: will the AI cloud ever earn software-like returns, or is Oracle trading high-margin software dollars for lower-margin rent? Management says infrastructure margins "improve rapidly" once data centers reach full contracted revenue, targeting a 30–40% operating margin for OCI. That is the promise the entire valuation rests on.

3. Returns on capital — the heart of the debate

A great business earns high returns on the capital it employs. Oracle's traditional software did exactly that. But as the company pivots to capital-intensive infrastructure, return on invested capital has been cut in half — from ~18% in FY2021 to ~8.7% in FY2026 — even as revenue accelerates. Growth is not free; it is being bought with a balance sheet.

Source: Derived from Oracle reported financials, FY2021–FY2026. ROE is omitted as meaningless — buybacks drove book equity near zero, so reported ROE swings between +7,300% and -400%.

This matters because Oracle is deploying capital at unprecedented scale. The bet only works if incremental capital earns well above its ~5% blended cost of debt. At a 9% consolidated ROIC, the spread is thin; at the promised high-20s, it is enormous. Both cannot be true for long — the contracted backlog will resolve it.

4. The cash-flow regime change

Until two years ago Oracle was a textbook cash compounder, converting ~22% of revenue to free cash flow. Then capex went vertical — from $6.9B (FY2024) to $55.7B (FY2026), roughly 6× depreciation — and free cash flow swung from +$11.8B to -$23.7B. Operating cash flow is actually rising (the software engine is healthy); it is the buildout that turns the company cash-negative.

Source: Oracle reported cash-flow statements, FY2022–FY2026. Capex shown as gross cash outlay. FY2026 FCF is operating cash flow minus capex.

Watch one piece of financial engineering: management now headlines a "net cash outlay for capex" that subtracts customer prepayments and bring-your-own-hardware. For FY2027 it guides to ~$70B net but admits reported capex will be $20–25B higher. The prepayments are real and genuinely lower the funding need — but the metric exists to make a colossal number look smaller. Underwrite the gross figure.

The funding plan is explicit: ~$40B of new debt and equity in FY2027, including a $20B at-the-market equity issuance. Which brings us to the most striking capital-allocation reversal in the company's history.

From cannibal to issuer

For fifteen years Oracle was a share cannibal, retiring nearly half its shares and using leverage to do it — buybacks peaked at $36.6B in FY2019. That program is now effectively over (just $0.2B in FY2026), and the share count has begun rising again as the company issues stock to fund data centers. The engine that returned cash to owners has been redirected into the ground.

Source: Oracle cash-flow statements (repurchases) and period-end share counts, FY2019–FY2026.

The share count bottomed at 2.66B (FY2022) and has climbed back to 2.88B. For a value investor this is a regime change worth internalizing: the per-share compounding that powered Oracle's old returns is paused, and dilution is now a headwind rather than a buyback tailwind. The thesis must come from growth in intrinsic value, not from shrinking the share base.

5. The moat: what is durable, what is not proven

Not all of Oracle's competitive advantage is equal. The honest read is that the database is a fortress, the applications are sticky, and the AI cloud has no proven moat yet — it competes in a market where capacity, not loyalty, often wins the marginal deal.

Source: Synthesis of Oracle FY2025 10-K, FY2026 earnings commentary (multicloud +404%, GPU utilization 97.5%, renewal data), and competitive analysis.

The mechanism that ties it together is the database as a gravity well. Decades of mission-critical corporate data already live in Oracle databases. The pitch — "run your AI next to your data, on our full stack" — converts a 40-year-old moat into a reason to rent new infrastructure. That is the strategic logic of the buildout: defend the database by becoming the place AI workloads run. If it works, the moat extends; if AI workloads prove cloud-agnostic and price-driven, Oracle is renting commodity capacity with a balance sheet.

6. How Oracle stacks up economically

Against its named peers, Oracle is a genuine outlier: it pairs a large revenue base with accelerating growth — a combination none of the others has. But that growth now costs it on every quality metric. Its gross margin is the lowest in the group (the infrastructure drag), its FCF margin has gone negative while pure-software peers run 23–35%, and its ROIC has fallen below the better-quality names.

Source: Latest fiscal year per company (Oracle FY2026; MSFT/SAP/IBM/NOW FY2025; CRM/WDAY FY2026). SAP converted to USD. FCF margin = (operating cash flow − capex) ÷ revenue. ROIC from reported financials.

The picture: Oracle has chosen to spend its way to growth, sacrificing near-term cash and returns. Microsoft is the instructive contrast — it runs a similar full-stack-plus-hyperscaler model but does it at 45.6% operating margin, 25% FCF margin, and 21.8% ROIC, because its cloud is mature and self-funding. Oracle is making the same bet Microsoft already won, but earlier in the curve, smaller, and with far more leverage. The applications-only peers (CRM, NOW, WDAY) show what clean software economics look like — 30%+ FCF margins — and underline how much the infrastructure mix is costing Oracle today.

Source: Market caps mid-June 2026 (peer valuations file); EV/Revenue derived from reported net debt. Oracle at the current ~$184 share price.

The bubble chart delivers the punchline: the market already pays Oracle a Microsoft-like ~9× EV/sales — far above SAP, IBM, Salesforce and Workday — on the strength of the backlog. You are not buying Oracle cheap relative to its own growth; you are buying the expectation that the backlog converts at high returns.

7. How to value it — the lens

Do not value Oracle on a single multiple. Value the two engines separately and ask what the current price implies.

Source: Derived at the current ~$184 share price (June 18, 2026), ~2.88B shares; net debt ~$98B (gross debt $129.5B − cash $31.9B); EBITDA ~$29.9B FY2026. Stock is down ~18% from its ~$226 fiscal-year-end level.

The sum-of-the-parts frame:

Part A — the software annuity. ~$50B of high-margin, recurring database + applications + support revenue with ~30% operating margins and software-grade ROIC. On its own this is a defensive compounder worth a premium software multiple. It is the part of Oracle you can underwrite with confidence.

Part B — the AI-infrastructure call option. The $638B RPO, the $55B+ annual capex, the 30–40% margin promise, the high-20s project-ROIC claim. This is a leveraged bet on AI capital spending staying strong, mega-customers honoring contracts, and Oracle executing a hyperscaler buildout from a smaller base. It is worth a great deal if it converts — and a liability if it does not.

What the current price implies is that Part B already works. EV/EBITDA of 21× and EV/sales of 9× on a business with single-digit consolidated ROIC and deeply negative free cash flow can only be justified by the forward numbers — management's +34% FY2027 revenue, +31% revenue CAGR and +28% EPS CAGR through FY2030. On the FY2027 non-GAAP EPS guide of $8.05 the stock trades ~23× forward; on FY2030 targets it looks reasonable — if you believe them.

Source: Synthesis of FY2026 disclosures, FY2027 guidance, long-term targets, and Moody's-flagged concentration risk in ~$300B of recently signed AI contracts.

Bottom line

Oracle is a wonderful old business wrapped around a high-stakes new one. The database-and-applications annuity is among the highest-quality franchises in software — sticky, cash-rich, moated. But the company has staked its balance sheet on becoming the fourth hyperscaler, and that bet has cut consolidated returns on capital in half, turned free cash flow sharply negative, ended the buyback era, and started the dilution era. The market is already paying for the bet to succeed. For an intelligent investor the question is not "is Oracle a good business?" — the annuity clearly is — but "am I being paid to take the conversion risk on $638B of backlog financed with $130B of debt and new equity?" Underwrite the two engines separately, demand evidence of the free-cash-flow inflection and the margin recovery, and remember that the per-share compounding that made Oracle great is, for now, switched off.