ORCLOracle Corporation

ORCL — dossier, 2026-09-23

Slot names-ai-capex-4. Drafted by Claude in an unattended cloud routine. Nothing here is human-verified. Every figure carries the document it was read from; anything from background knowledge is marked [background]. Quotes were located by substring search in the source and copied, never re-typed. Never edited after today.

Price reference: $149.20 close, 2026-09-22, volume 31,194,188 (ROIC.ai NYSE:ORCL).

---

The one-line version

Oracle's income statement looks better than ever — revenue +29.6%, operating margin 34.8%, net income +62.6% — and in the same quarter it spent $28.5 billion of capex, took $11.4 billion of customer prepayments that its own accounting calls financing, sold its entire $20 billion ATM in three months, and bought back nothing. The 2026-09-20 cluster file called Oracle the first large spender to cross into sustained negative free cash flow and stop buybacks. This quarter's primary documents confirm both and add the funding mechanism: the customers are paying for the buildout in advance, and Oracle discloses it as a significant financing component.

---

S1 — Source

watchlist (owner, Apple Stocks, 2026-09-20), row: {ticker: "ORCL", label: "Oracle Corporation", coverage: sec_domestic, cluster: ai_datacenter, cik: 1341439}. No Tier 0 promotion.

S2 — Kill test

From XBRL company facts (data.sec.gov, User-Agent sent) and the two filings. Oracle's fiscal year ends 31 May, so FY2026 = year to 2026-05-31 and Q1 FY2027 = quarter to 2026-08-31.

The balance sheet in three months (2026-05-31 → 2026-08-31): total assets $261,759m → $303,259m; PP&E net $99,957m → $127,845m; cash $31,289m → $36,369m; current deferred revenue $9,916m → $14,686m; other non-current liabilities $16,178m → $28,183m; total debt (current + non-current notes payable) $129,541m → $125,337m; Oracle stockholders' equity → $66,772m; common shares outstanding 2,880m → 3,024m.

Survives-or-dies. Oracle does not die on the income statement — it is compounding revenue at 29.6% with an expanding operating margin. It dies, if it dies, on the funding of the build and on the share count that funding produces. Read the quarter as cash: operating cash flow $23,103m minus capex $28,499m is −$5.4bn of free cash flow; strip out the $11.4bn of customer prepayments that sit inside operating cash flow and it is −$16.8bn in a single quarter. That gap was closed with $19.9bn of equity sold through an at-the-market programme that was exhausted inside the quarter, and Oracle bought back nothing while paying $1,565m of dividends. Net debt is ~$88.3bn. There is no solvency question at this size and this margin; there is a dilution question, and it is live.

The share-count arithmetic. 2,880m shares at 2026-05-31 → 3,024m at 2026-08-31 is +5.0% in three months, all of it the ATM. The $20bn authorisation granted on 2026-02-02 lasted two quarters. The $4,954m of 6.50% Series D mandatory convertible preferred converts on top. If the FY2027/FY2028 "peak capex" years run at anything like the Q1 FY2027 rate, the question is not whether Oracle raises again but how.

S3 — Filings, gate-minimum

Read: FY2026 10-K (2026-06-22) risk factors and MD&A; Q1 FY2027 10-Q (2026-09-11) MD&A delta, Notes 1 and 7, Item 5; the Q1 FY2027 call (turns 7–37 of 37); Form 4 cadence and Form 144 primary documents via 3spread and EDGAR. Not read: the DEF 14A (2025-09-26), the FY2025 10-K, older 10-Qs, any transcript before Q1 FY2027, and the efts.sec.gov auditor/restatement sweep.

The three biggest risks management itself discloses

1. "If we are unable to secure data center capacity at affordable rates or do not accurately plan for and manage our infrastructure capacity requirements, our profitability may decline." — 10-K FY2026, Risk Factors, heading. Three things inside it matter more than the heading:
2. "In certain OCI offerings, we are more concentrated among a number of large customers, which could increase these risks." — the concentration admission, with no customer named anywhere.
3. "if we overestimate customer demand or any of our key customers are unable to pay or otherwise perform under their contracts with us, we could be locked into multi-year commitments for excess data center space and related capital expenditures, as well as associated financings, without receiving corresponding revenue."
4. "the terms, renewal options and pricing adjustments in our long-term data center leases typically do not align with the duration and pricing of customer contracts" — the maturity mismatch, stated by the issuer.
5. "There are risks associated with our outstanding and future indebtedness." — 10-K FY2026, Risk Factors, heading. The body names refinancing risk, rating actions, and that a downgrade could "affect the terms or availability of certain long-term commitments (including data center leases)" and increase collateral and credit-support requirements. On $125.3bn of debt against a build funded partly by counterparties' credit, that is not boilerplate.
6. Supply terms as a disclosed concession. "industry supply capacity for AI accelerators, including graphics processing units, as well as memory devices, is competitive, and we at times have to accept less favorable terms with suppliers to minimize supply constraints" — with the consequence stated: increased excess and obsolescence risk, and operating costs for capacity that is not yet revenue-generating.

The number that reframes the name: $664 billion of RPO

"Remaining performance obligations were $ 664 billion as of August 31, 2026 , of which we expect to recognize approximately 13 % as revenues over the next twelve months" — with 37% in months 13–36, 34% in months 37–60 and the remainder thereafter (10-Q Note 1).

Against FY2026 revenue of $67.4bn that is roughly ten years of sales, booked and not delivered, and 87% of it is more than twelve months out. Oracle has never disclosed which customers it belongs to. On the call an analyst asked directly whether the RPO growth comes from AI labs, semiconductor companies or sovereigns; the answer was "really broad-based" (see Dodges). The 10-K's own concentration language sits next to it.

Where the cash came from: the customer financed the seller

"During the first quarter of fiscal 2027, we received $ 11.4 billion of prepayments from customers that included a significant financing component. No prepayments from customers that included a significant financing component were received during the first quarter of fiscal 2026." (10-Q Note 1.) Oracle recognises interest expense on these separately from revenue, and discounts them at a rate reflecting "the credit characteristics of the party receiving financing" — i.e. Oracle's own.

This is the third distinct shape of the routine's standing cross-cutting finding about who funds whom in this complex:

• names-ai-capex-1: sellers underwrite buyers — NVDA's $105bn capped guarantee, AMD's penny warrants to OpenAI and Meta, AVGO co-sponsoring XPV with Apollo and Blackstone.
• names-ai-capex-3 (CBRS): buyers finance sellers — OpenAI's ~$1.0bn loan to Cerebras at 6% against the MRA, plus a penny warrant amortised as contra-revenue.
• Here: customers pre-fund the seller's capex, and the seller books it as financing. An executive laid out all three mechanisms on the call in one answer: supplier financing that lets Oracle "pay for the capacity as the customers pay us"; the customer buying the hardware outright and renting Oracle's operations around it ("bring your own hardware"); and the customer prepaying. His framing of the distinction was precise and worth keeping: "I didn't say, and I don't think myself nor Hillary said that it doesn't require additional CapEx. We said it doesn't require additional cash from Oracle."

Capex still happens. It is simply not on Oracle's cash flow statement, which is what "capex-light growth" means in practice for this name.

Transcript — Q1 FY2027 call, 2026-09-10 (read Q&A-first, SKILL §3.1)

Turns 7–37 of 37. Turns 1–6 (safe-harbor and prepared remarks) not read — one transcript call at this depth. Six analyst questions: Deutsche Bank, Mizuho, Barclays, Bernstein, Guggenheim, Jefferies, Evercore ISI.

A note on attribution. The Equibles transcript resolves the CFO (Hilary Maxson) and the IR head (Ken Bond) but leaves the two executives who answer the infrastructure and applications questions as "Speaker 9" and "Speaker 5". The connector's documentation says identities appear only when resolution is trusted. The four answers that decide this name are "Speaker 9"'s. No name is guessed here or in relationships.yaml.

Guidance and changes.

• FY2027 revenue guide raised to "at least $90 billion" (stated in the Mizuho question and not corrected) — against FY2026's $67,357m, ~+34%.
• FY2027 and FY2028 are "peak CapEx years" (stated in the Deutsche Bank question, not corrected).
• 850 megawatts delivered in Q1, and explicitly not from Shackelford, New Mexico, Wisconsin or Michigan — the four sites the market watches.
• Gross margin stepped down this quarter, as pre-guided at Q4; expected to "flatten" as the ramp completes; no new guidance until the October Investor Day.
• Renewal pricing: "when we go out and we have capacity up for renewal, that we actually can achieve higher prices to the order of 20%." Component cost inflation is expected to pass through with no gross-margin impact.
• Next earnings: 2026-12-14 (IR, closing remarks). Ken Bond is retiring; this was his last call.

What the analysts asked about. Five of the seven questions were about the same thing from different angles: how is this being paid for, and by whom. Zelnick on capex beyond FY2028 and on free-cash-flow timing; Panigrai on New Mexico and Wisconsin delivery risk to the FY2027 guide; Lenschau on component pricing and gross margin; Modler on which customers drive the capex-light RPO; DeFucci on gross margin, twice. Only Thill (SaaS) and Materne (AI data platform) asked about the product.

Dodges.

• Free-cash-flow timing. Zelnick asked when Oracle returns to positive free cash flow. The executive answering opened "I think there's like four questions in there, so I'm going to answer the parts that I want to answer, and then I'll make Hillary do the hard work", answered the funding-model part, and handed over. The CFO: "We haven't given a particular time frame on that yet, and we don't expect to give that." She then argued each project converts at roughly 100% of post-tax EBITDA once ramped, so the business is "somewhat quote self-funding at some point" — which is a claim about project economics, not an answer about consolidated free cash flow.
• RPO composition. Modler asked whether the capex-light RPO growth is AI labs, semiconductor companies or sovereigns. The answer corrected the premise, explained the three funding mechanisms, and said "it's really broad-based right … it doesn't matter if it's a startup or … the most valuable investment grade companies." No category, no concentration number, on a $664bn backlog the 10-K says is concentrated.
• Gross margin direction. DeFucci asked how to think about gross margin going forward; the answer was a case for watching operating margin instead. Pushed a second time — "Should gross margins continue to go down now from here, or should they be steady?" — the CFO gave "you can reasonably expect that gross margin would flatten" and deferred: "we haven't given any particular guidance today. We'll speak more about some of this stuff on the upcoming Investor Day."

Two of the three dodges point at the same October event. That makes the Investor Day the highest- information date on this name's calendar, and it falls before the 2026-12-14 10-Q.

What was answered well, and is checkable. The New Mexico / Wisconsin answer was specific and falsifiable: New Mexico construction on track, air permit in process, Bloom fuel cells for on-site generation; Wisconsin grid-delivered with the Public Service Commission, ATC and We Energies, "constantly evolving" energy design; and the flat claim that neither site affects FY2027 revenue or earnings guidance. Those are three statements a later filing can contradict.

S5 — Ownership

13F register at 2026-06-30 (3spread, CUSIP 68389X105, every row carrying its SEC filing URL; 2,513 filers against 3,419 the prior quarter — ingestion is incomplete and no exit is read from an absence):

Like INTC, the quarter's marginal buyers are dealers and multi-strategy funds rather than long-only managers. Jupiter Topco LLC appearing new at 12.7m shares was not identified in this pass and is flagged for a later look.

The insider item. Item 5 of the 10-Q: "Lawrence J. Ellison , our Executive Chair of the Board of Directors and Chief Technology Officer , adopted a new trading plan on June 22, 2026 . Mr. Ellison's plan is scheduled to terminate on October 24, 2026 … The trading plan is intended to permit Mr. Ellison to sell up to 50 million shares of Oracle common stock." At the 2026-09-22 close that is ~$7.5bn, and it runs out four days before the plan window closes on 2026-10-24.

Against that, the Form 144s 3spread holds for ORCL since 2026-03-07 are four and small: $66,064,000 proposed on 2026-06-24 (Director/Officer), $2,576,855.73 on 2026-06-23 (Officer), $1,522,827.08 on 2026-09-16 (Officer/Director), and one on 2026-04-16. On this evidence the 50m-share plan had not been executed at scale by 2026-09-16 — but Form 144 coverage on this source is not guaranteed complete, and the ownership denominator (Ellison's total holding) was not read because the DEF 14A was not opened. Both caveats belong with the figure.

• Form 4 cadence: 46 filings since 2026-03-07, a cluster of eleven on 2026-09-18 (a week after the 10-Q, i.e. the post-earnings window).
• Short interest (FINRA, via Equibles): 36,356,379 (2026-05-15) → 41,966,703 (2026-06-30) → 50,143,917 (2026-07-31) → 47,767,442 (2026-08-14) → 44,597,874 (2026-08-31); days to cover 1.6 → 2.3. It built 38% into midsummer and has eased since. Under 1.5% of 3,024m shares — small, and the shape matches the capex disclosures rather than the price. Equibles also returns a model estimate for the 2026-09-15 settlement (~47.8m, +7.1%, P10–P90 41.7–55.1m); it is an estimate, not a FINRA figure, and is not used here.
• Shelf / ATM / buyback: the $20bn ATM (2026-02-02, amended 2026-06-23) is exhausted — 141m shares for $19.9bn net. ~$6.3bn of buyback authorisation remains and nothing was repurchased in the quarter (0.4m shares / $93m a year earlier). Dividends continue at $0.50 per common share quarterly plus $1,625 per preferred share. Oracle is issuing equity and paying a dividend at the same time.

S8 — Valuation and sizing ⏳ OPEN (user)

Today. $149.20 (2026-09-22). 3,024m common shares → ~$451bn of market value; plus $4,954m of preferred and ~$88.3bn of net debt → ~$544bn enterprise value. FY2026 net income $17,087m puts the trailing P/E near 26x; on the Q1 FY2027 run rate it is nearer 24x.

Out-year model. Assumptions, all stated: FY2027 revenue $90bn (management's own raised guide), FY2028 $115bn (+28%, slower than FY2027's +34%); operating margin held at 31%, i.e. below the 34.8% Q1 FY2027 printed, because the ramp adds depreciation faster than revenue; net interest $8bn on $125bn of debt growing with the build; an 18% tax rate; and 3,150m shares, which assumes the Series D converts and roughly one further $20bn-scale equity raise. Preferred dividends $324m.

What that says. Unlike INTC, ORCL's base case lands essentially on the traded price. The multiple is not the argument here; the share count is. Every further $20bn of ATM at ~$150 is ~133m shares, ~4.4% dilution — and Oracle consumed a $20bn authorisation in a single quarter. Two more of those between here and FY2028 take the base case from ~$142 to roughly $130 with nothing else changing. The model's sensitivity is entirely to how the peak-capex years get funded, which is exactly the question the CFO declined to timeframe.

The bear case is not a multiple compression; it is the 10-K's own sentence about being locked into multi-year commitments and associated financings without corresponding revenue, applied to a $664bn backlog whose counterparties are undisclosed.

Buy price and size are owner-only (skill §5). Liquidity: 31.2m shares on 2026-09-22, ~$4.7bn.

S9 — Monitoring

• Next earnings: 2026-12-14, stated by IR in the call's closing remarks. Before that, the October 2026 Investor Day, to which the CFO twice deferred the gross-margin and free-cash-flow questions.
• Feed: Tier 0 EDGAR sweep covers cik 1341439.
• Leading indicator between quarters: the prepayment line in Note 1 — $11.4bn this quarter, $0 a year ago. It is disclosed quarterly, it is roughly half of operating cash flow, and it is the clearest single measure of whether customers are still willing to fund the build. Read it beside RPO ($664bn) and the next-twelve-months conversion percentage (13%).
• Trigger that converts this to a buy: capex peaking on the stated FY2027–FY2028 schedule with the prepayment line holding and no second ATM — i.e. the funding model proving out before the free-cash-flow crossover. A date for positive free cash flow given at the October Investor Day would be the single cleanest confirmation available, because management has so far refused to give one.
• Trigger in the other direction: a new ATM authorisation, or RPO flat-to-down in the 2026-12-14 10-Q, or the next-twelve-months conversion percentage falling below 13%.

---

Access and evidence limits, this run

• DEF 14A not read (2025-09-26). No insider ownership percentage, no compensation structure, no denominator for the 50-million-share Ellison plan. Largest gap; it is the scorecard's next action.
• One transcript, not two. Q1 FY2027 only, turns 7–37; prepared remarks not read.
• Two speakers unresolved in that transcript. The executive who answered the four questions that decide this name is "Speaker 9" to the connector, and this dossier does not guess who that is.
• No efts.sec.gov full-text sweep for auditor change or restatement language.
• No GetCustomerConcentration / GetGoingConcernStatus / GetExecutiveChanges calls — Equibles budget. Customer concentration would have been the most valuable of the three and is the one place the budget bit; the 10-K's qualitative admission is what stands in for it.
• Form 144 coverage may be incomplete on 3spread; the four filings found are treated as a floor, not a census.
• Jupiter Topco LLC not identified — new at 12.7m shares in the quarter.
• The out-year model is the routine's own, on stated assumptions. FY2027 revenue is management's guide; everything from FY2028 is not.