The AI infrastructure backbone hiding in plain sight
While everyone watched OpenAI, Oracle quietly became the cloud backbone that enterprises actually run their AI on. $553B backlog and growing.
No Friday close is on the record for ORCL yet. The weekly job prices the covered universe; a name it cannot price is listed as missing rather than estimated.
| Date | Type | What happened | Source |
|---|---|---|---|
| 2026-09-11 | editorial note | Capex figure revised: $90 → $28,499MCapex moved from $90 (deep-dive-2026-07-25-refresh.md) to $28,499M (deep-dive-2026-09-11.md). | dossier |
| 2026-09-11 | editorial note |
| Revenue figure revised: $18,101M → $19.3BRevenue moved from $18,101M (deep-dive-2026-07-25-refresh.md) to $19.3B (deep-dive-2026-09-11.md). |
| dossier |
| 2026-09-11 | editorial note | Verdict changed: none (no verdict, no call this pass) → The Q1 FY27 print settles the two questions the July dossier could not — and both settle against the received narrative in opposite directions.…Before (deep-dive-2026-07-25-refresh.md): none (no verdict, no call this pass) After (deep-dive-2026-09-11.md): The Q1 FY27 print settles the two questions the July dossier could not — and both settle against the received narrative in opposite directions. The backlog is NOT lengthening (the near-term conversion share has ticked UP for three straight quarters; the lengthening was a single step-change in Q1 FY26 when the mega-contracts landed, not a trend), while the FUNDING has decisively changed character: $11.4B of the record $23.1B operating cash flow is customer credit Oracle itself books interest expense on, and capex at 1.48x revenue runs 3-5.5x every hyperscaler even after the finance-lease adjustment that usually closes such gaps. Oracle is not out-spending its peers as a share of revenue; it is spending a different KIND of money. | dossier |
| 2026-07-25 | editorial note | Capex figure revised: $55.7B → $90Capex moved from $55.7B (deep-dive-2026-06-23.md) to $90 (deep-dive-2026-07-25-refresh.md). | dossier |
| 2026-07-25 | editorial note | Margin figure revised: 59% → 63%Margin moved from 59% (deep-dive-2026-06-23.md) to 63% (deep-dive-2026-07-25-refresh.md). | dossier |
| 2026-07-25 | editorial note | Revenue figure revised: $67,357M → $18,101MRevenue moved from $67,357M (deep-dive-2026-06-23.md) to $18,101M (deep-dive-2026-07-25-refresh.md). | dossier |
| 2026-07-25 | editorial note | Verdict changed: The FY26 10-K turns the bull narrative into audited fact and the bear narrative into a footnote you can now read — $67.4B revenue, $638B RPO, but only 12% converts inside a year, −$2…Before (deep-dive-2026-06-23.md): The FY26 10-K turns the bull narrative into audited fact and the bear narrative into a footnote you can now read — $67.4B revenue, $638B RPO, but only 12% converts inside a year, −$23.7B FCF, $129.5B debt, a 13% workforce cut, and a $19B post-quarter purchase commitment; good news kept beating and the stock kept falling (~$166, −52% from peak), which is the whole thesis in one line. After (deep-dive-2026-07-25-refresh.md): none (no verdict, no call this pass) | dossier |
| 2026-06-23 | editorial note | Capex figure revised: $70B → $55.7BCapex moved from $70B (deep-dive-2026-06-18.md) to $55.7B (deep-dive-2026-06-23.md). | dossier |
| 2026-06-23 | editorial note | Margin figure revised: $150B → 59%Margin moved from $150B (deep-dive-2026-06-18.md) to 59% (deep-dive-2026-06-23.md). | dossier |
| 2026-06-23 | editorial note | Revenue figure revised: $19.4B → $67,357MRevenue moved from $19.4B (deep-dive-2026-06-18.md) to $67,357M (deep-dive-2026-06-23.md). | dossier |
| 2026-06-23 | editorial note | Verdict changed: A 49-year-old database company has reinvented itself as the lender-of-last-resort GPU landlord — $638B of backlog is real and mostly OpenAI, and the entire thesis now rides on whethe…Before (deep-dive-2026-06-18.md): A 49-year-old database company has reinvented itself as the lender-of-last-resort GPU landlord — $638B of backlog is real and mostly OpenAI, and the entire thesis now rides on whether one un-IPO'd customer can pay a bill Oracle is borrowing $100B+ to build for. After (deep-dive-2026-06-23.md): The FY26 10-K turns the bull narrative into audited fact and the bear narrative into a footnote you can now read — $67.4B revenue, $638B RPO, but only 12% converts inside a year, −$23.7B FCF, $129.5B debt, a 13% workforce cut, and a $19B post-quarter purchase commitment; good news kept beating and the stock kept falling (~$166, −52% from peak), which is the whole thesis in one line. | dossier |
The verdict
The Q1 FY27 print settles the two questions the July dossier could not — and both settle against the received narrative in opposite directions. The backlog is NOT lengthening (the near-term conversion share has ticked UP for three straight quarters; the lengthening was a single step-change in Q1 FY26 when the mega-contracts landed, not a trend), while the FUNDING has decisively changed character: $11.4B of the record $23.1B operating cash flow is customer credit Oracle itself books interest expense on, and capex at 1.48x revenue runs 3-5.5x every hyperscaler even after the finance-lease adjustment that usually closes such gaps. Oracle is not out-spending its peers as a share of revenue; it is spending a different KIND of money.
Primary sources
SEC filings
Source documents — open to read in full
Q1 FY27 printed, and it beat. Revenue $19.3B, +30%; cloud $11.6B, +62%; IaaS $7.4B, +121%; SaaS $4.2B, +10%; software $5.5B, −3%; GAAP EPS $1.56, +55%; non-GAAP EPS $1.92, +30%; GAAP operating income $6.7B, +57%; operating cash flow $23,103M, +184%; Free cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. −$5,396M; Capital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. $28,499M. Remaining performance obligationsRevenue a company has already signed contracts for but has not yet delivered or booked. $664B, up $209B y/y and +$26B q/q. ``
THE CORRECTION THAT MATTERS — the backlog is not lengthening. The July dossier and the companion note both carried a "lengthening backlog" reading, sourced to a 33% → 12% fall in the twelve-month conversion share. Reading the full Form 10-QThe quarterly version of the annual report. Lighter, and not audited. series now on the shelf shows that framing is wrong. See Lens 5. The near-term share went 33% → 10% in a single quarter (Q1 FY26, when the mega-contracts landed) and has since gone 10 → 10 → 12 → 12. It is flat-to-rising, not falling. The mix shifted once, at a known event, and stopped.
The funding composition changed character, and the quarterly ladder is now visible. Customer prepayments carrying a significant financing component — Oracle's own accounting language for money on which it books interest expense — ran $0 · $0 · $0 · $4,592M across FY26's four quarters and $11,363M in Q1 FY27 alone. `` That is 2.5x the entire prior fiscal year in ninety days, and it is 49% of the record operating cash flow.
THE OPEN ITEM IS NOW CLOSED — the hyperscaler capex comparison is done, finance-lease adjustment explicit. See Lens 7. Oracle's capex/revenue is 1.48x against 0.27–0.50x for MSFT, GOOGL, AMZN and META, and adding finance-lease right-of-use additions — the adjustment that normally narrows such gaps, because the hyperscalers lease what Oracle buys — does not close it.
The stock recovered a third off the low. $114.99 (2026-07-24 close, prior dossier) → $152.94 (2026-09-10 close), +33.0%. `` The print landed after that close; the 2026-09-11 session had not opened at time of writing, so the market's verdict on Q1 FY27 is not in this dossier.
The balance sheet took the equity. Stockholders' equity $43,056M → $67,196M (the $19,909M net ATM); total debt $129.5B → $125,337M (notes payable current $7,625M + non-current $117,712M) on $4,202M of senior-note repayment; total assets $303,259M. ``
Oracle volunteered a sentence nobody asked for. "Based on the structuring of those new contracts, the Company confirms there is no incremental impact on its plans to raise capital." `` Carried forward from the companion note because it is the single most defensive line in the document and it is management's own framing of the question this dossier is asking.
| Line | Q1 FY27 | Q1 FY26 | Change |
|---|---|---|---|
| Total revenue | $19.3B | — | +30% |
| Cloud (IaaS+SaaS) | $11.6B | — | +62% |
| — Cloud infrastructure (IaaS) | $7.4B | — | +121% |
| — Cloud applications (SaaS) | $4.2B | — | +10% |
| Software | $5.5B | — | −3% |
| Services | $1.4B | — | +5% |
| Hardware | $0.8B | — | +15% |
| GAAP operating income | $6.7B | — | +57% |
| GAAP EPS | $1.56 | — | +55% |
| Non-GAAP EPS | $1.92 | — | +30% |
| Operating cash flow | $23,103M | $8,140M | +184% |
| Capital expenditures | $28,499M | $8,502M | +235% |
| Free cash flow | −$5,396M | — | — |
| RPO | $664B | $455B | +$209B |
`` — every line read from the press release and its financial statements, not from coverage.
Guidance: FY27 revenue at least $90B, non-GAAP EPS $8.10. Q2 revenue +30–34%, cloud +65–71% USD, non-GAAP EPS $1.85–$1.93 USD. Note the footnote Oracle buried and most coverage dropped: including the Q2 FY26 one-time Ampere gain, guided Q2 FY27 non-GAAP EPS is a decline of 14–18%. `` The "+21% to +25%" headline is an ex-item number.
Dividend: $0.50/share declared, record 2026-10-09. ``
The July dossier, the June dossier and the companion research note all carried some version of "the backlog is lengthening", anchored on a fall from ~33% to 12% in the share of RPO converting within twelve months. Six 10-Qs were ingested to this shelf during this run. Read in sequence, they say something different:
| As of | Filing | RPO | 1–12m | 13–36m | 37–60m | Remainder |
|---|---|---|---|---|---|---|
| 2025-05-31 | FY25 Form 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. | $137.8B | 33% | 41% | 23% | ~3% |
| 2025-08-31 | Q1 FY26 10-Q | $455.3B | 10% | 25% | 34% | ~31% |
| 2025-11-30 | Q2 FY26 10-Q | $523.3B | 10% | 30% | 35% | ~25% |
| 2026-02-28 | Q3 FY26 10-Q | $552.6B | 12% | 31% | 35% | ~22% |
| 2026-05-31 | FY26 10-K | $638B | 12% | 34% | 34% | ~20% |
``
Three things follow and none of them is in the coverage:
1. The lengthening was a single step-change, not a trend. RPO went $137.8B → $455.3B in one quarter (+230%) and the duration profile reset in that same quarter. That is not a backlog quietly stretching out; it is a small near-dated book being swamped by an enormous long-dated one in a single bookings event. The mix moved because the numerator moved.
2. Since that reset, the curve has been shortening at the margin, for three consecutive quarters. 10% → 10% → 12% → 12% on the near-term share, and 25% → 30% → 31% → 34% on months 13–36, while the beyond-60-month remainder fell 31% → 25% → 22% → 20%. Every one of those moves is in the direction of the book converting sooner, not later. Anyone still telling you the Oracle backlog is lengthening is describing a quarter in mid-2025.
3. The dollars, not the percentages, are the story. The percentage is a ratio against a base that quadrupled. In dollars:
| As of | 1–12m share | Implied $ | `` |
|---|---|---|---|
| 2025-05-31 | 33% | ~$45B | 0.33 × $137.8B |
| 2025-08-31 | 10% | ~$46B | 0.10 × $455.3B |
| 2026-02-28 | 12% | ~$66B | 0.12 × $552.6B |
| 2026-05-31 | 12% | ~$77B | 0.12 × $638B |
The near-dated book barely moved through the mega-contract quarter (~$45B → ~$46B) and has grown ~67% since. The mega-contracts genuinely did add almost nothing to the next twelve months when signed, and have been pulling forward since.
What this does to the FY27 guidance fit, restated. 12% of $638B ≈ $77B of contracted conversion against "at least $90B" guided — about 85% of guided revenue is contracted backlog. `` That is a tight fit rather than a cushion, and it is unchanged from the companion note's reading. What is new is that the curve is not deteriorating beneath it.
The load-bearing year is still FY2028. The 34% in months 13–36, spread over two years, is ~$108B/yr `` against ~$77B in the current year. Whether that step-up arrives is the question the whole file turns on, and it does not resolve until roughly June 2027.
Increase in deferred revenues from customer
prepayments with significant financing component... $11,363M (Q1 FY26: $—)
Increase in other deferred revenues................... 3,997 (Q1 FY26: 2,550)
Net cash provided by operating activities............. 23,103 (Q1 FY26: 8,140)
Capital expenditures.................................. ( 28,499 ) (Q1 FY26: 8,502)
Proceeds from ATM common stock issuance, net.......... 19,909 (Q1 FY26: —)
(Repayments of) short-term financing re: capex, net... ( 830 ) (Q1 FY26: 1,958)
Repayments of senior notes / term loans / other....... ( 4,202 ) (Q1 FY26: 1,052)
Payments of dividends................................. ( 1,565 )
``
And Oracle's own reconciliation, which gives the quarterly ladder the July dossier could not see:
| Q1 FY26 | Q2 FY26 | Q3 FY26 | Q4 FY26 | FY26 | Q1 FY27 | |
|---|---|---|---|---|---|---|
| Capital expenditures | $8,502 | $12,033 | $18,635 | $16,493 | $55,663 | $28,499 |
| Less: other short-term financing re: capex | (1,958) | 760 | (1,469) | (678) | (3,345) | 830 |
| Less: customer prepayments w/ financing component | — | — | — | (4,592) | (4,592) | (11,363) |
| Net cash outlay for capex | $6,544 | $12,793 | $17,166 | $11,223 | — | $17,966 |
arithmetic on three primary lines]
Read the bottom row across. Capex tripled year-on-year while the net cash outlay grew 175% — the wedge is entirely the prepayment line, which did not exist a year ago. Oracle is spending three times as much and putting up less than two times as much of its own cash, because customers are now funding the gap. That is either the best possible evidence that demand is real (customers pre-paying for capacity they cannot get elsewhere) or the mechanism by which a funding constraint is deferred rather than solved. The filings do not distinguish between those two readings, and anyone asserting either one is asserting.
Stripping the financing-component prepayments, operating cash flow is ~$11.7B, +44% — good, not a record, and free cash flow on that view is **~−$16.8B** rather than −$5.4B. Neither presentation is wrong; GAAP does put prepayments in operating. But the "+184%, record" headline is carried by a line item that was zero in the comparative quarter, and the composition is disclosed one line above the total.
The Q1 FY27 call was held 2026-09-10, 4:00pm Central. No transcript could be obtained from this environment. Oracle's own IR pages (investor.oracle.com, oracle.com/investor/) return 403/404 to programmatic fetch, and the transcript hosts the skill names as reliable (Motley Fool, Insider Monkey) plus Benzinga and AlphaStreet all returned 403 in the companion note's attempt the same day. transcripts/ for this company remains empty.
Two figures circulating in the coverage live only on that call and are not in any Oracle filing: 97.9% GPU utilisation and a 20% premium on renewal and resale contracts. They are absent from Ex-99.1 — checked directly. They are recorded here as unverified management commentary and must not be repeated as facts. The renewal premium in particular is the single cleanest indicator of whether the capacity is genuinely scarce, which makes its unverifiability more costly than its size suggests.
Sentiment trend across the last 3–4 calls: carried forward from the previous dossier, unrefreshed.
The July dossier and the companion note both deferred this deliberately, on the correct grounds that forcing the comparison would produce a number that looked rigorous and was not. It is done here from SEC XBRL companyconcept facts — the same numbers the companies tagged in their own 10-Q/10-K — with the finance-lease adjustment made explicit, which is what the deferral was waiting for.
Latest reported quarter, each company on its own fiscal calendar:
| Period (quarter ended) | Capex | Finance-lease ROU additions | Capex + FL | Revenue | Capex/rev | (Capex+FL)/rev | |
|---|---|---|---|---|---|---|---|
| ORCL | 2026-08-31 (Q1 FY27) | $28,499M | not available | — | $19,300M | 1.48x | — |
| MSFT | 2026-06-30 (Q4 FY26) | $35,802M | $5,122M | $40,924M | $90,007M | 0.40x | 0.45x |
| GOOGL | 2026-06-30 (Q2 26) | $44,924M | $691M | $45,615M | $119,796M | 0.37x | 0.38x |
| AMZN | 2026-06-30 (Q2 26) | $54,208M | $563M | $54,771M | $200,606M | 0.27x | 0.27x |
| META | 2026-06-30 (Q2 26) | $30,116M | not tagged quarterly | — | $60,801M | 0.50x | — |
Two figures are deliberately blank.
What the table says.
Oracle's capex intensity is 3.0x to 5.5x every hyperscaler's, on the same quarter, on comparably tagged facts. Against cloud revenue alone ($11.6B) it is 2.46x ``.
The standard objection to exactly this comparison is that the hyperscalers lease a large share of their datacenter capacity — so their purchased-capex understates their true build and the gap is an artifact. The finance-lease adjustment is in the table, and it does not rescue the objection. It is material only for Microsoft, moving it from 0.40x to 0.45x — and 0.45x is still under a third of Oracle's 1.48x. For Google and Amazon it is rounding. Even crediting Oracle with a full $1.5B of unrecorded finance leases would leave it at ~1.56x, moving it further away, not closer.
The honest limit of this table. Capex/revenue is not a unit cost and not an efficiency measure. Oracle is at an earlier point in a build cycle than the other four, spending against contracted revenue that has not converted yet; the hyperscalers are spending against revenue already recognised. A high ratio at this point in a build is what a build looks like. What the table establishes is magnitude, not judgment — Oracle is running an intensity no hyperscaler is running, and the usual accounting explanation for such a gap is not the explanation here. Whether that is aggression rewarded or aggression punished is the question, not the answer.
One more denominator Oracle gives that nobody else does: 850MW delivered in Q1, so ~$33.5M of capex per MW delivered ``. This conflates spend-in-period with capacity-delivered-in-period and is not a unit cost; no peer publishes MW added, so it has no comparator and is recorded only so a future quarter has something to move against.
Valuation multiples (P/E, EV/Sales, EV/EBIT, dividend yield, 5-yr ROE) for the peer set: n/a this pass. No market-data source that works from this environment returned multiples, and the skill's rule is n/a over a plausible-looking fabrication. The July dossier's ORCL multiples (fwd ~14.3x, trailing ~19.7x at $114.99) are stale by a 33% price move and are not carried forward.
| Date | Close | Note |
|---|---|---|
| 2026-06-23 | $166.27 | prior-prior dossier `` |
| 2026-07-24 | $114.99 | prior dossier; ~52-wk low $114.75 `` |
| 2026-09-03 | $154.04 | `` |
| 2026-09-04 | $158.78 | |
| 2026-09-08 | $162.52 | intraday high $170.70 |
| 2026-09-09 | $161.63 | |
| 2026-09-10 | $152.94 | print day — results released AFTER this close |
``
Two observations, and one deliberate silence.
The recovery preceded the print. From $114.99 to $170.70 intraday on 2026-09-08 is +48.5% in six weeks with no filing and no earnings in between — the same six-week window in which the July dossier recorded a fresh 52-week low, an S&P downgrade to BBB− and a widening projected cash deficit. Whatever re-rated the stock was not a disclosure. That fits the pattern the June dossier named: for this name the tape and the fundamentals have been decoupled in both directions, and the decoupling itself is the regime signal.
Print day fell 5.9% from the prior close, before the results. $162.52 → $152.94 across 2026-09-09/10 on the heaviest volume of the window (27.8M shares on 2026-09-10). Positioning moved ahead of the number.
The silence: the 2026-09-11 US session had not opened when this was written. The market's actual verdict on Q1 FY27 is not in this dossier and must not be inferred from anything above. That is the single most obvious thing a reader will want and it is the one thing here that would be invented.
the previous dossier. No management change, no capital-allocation regime change beyond the ATM already recorded. The $19.9B equity issuance is the most consequential capital-allocation act of the period and is judged in Lens 10 and 12 rather than re-opening the archetype assessment.No new SEC enforcement. regulatory/regulatory-findings.md (fetched 2026-06-18) records zero Litigation Releases and zero AAERs naming Oracle in the five-year window. Not re-fetched this pass — the file is 85 days old, marginally outside the skill's 7-day freshness convention, and is flagged as such rather than silently relied on. The securities class action on the cloud narrative recorded in the June dossier is carried forward unchanged; no development was found in the primary documents read here.
The quality-of-earnings concerns, updated:
The financing-component prepayment is the headline flag, and it is Oracle's own characterisation, not an analyst's. The FY26 10-K policy note states the discount rate on such contracts is set "at a rate that reflects the credit characteristics of the party receiving financing" and that Oracle "recognize[s] interest expense related to significant financing components separately from revenue." `` The 10-K also states plainly: "During fiscal 2026, we received $4.6 billion of prepayments from customers that included a significant financing component. No prepayments were received from customers that included a significant financing component during fiscal 2025 and 2024." A funding channel that went $0 → $0 → $4.6B → $11.4B-in-one-quarter in six quarters is a material change in the economics of the business that appears nowhere in the income statement.
Non-GAAP framing did work this quarter. The guided Q2 EPS growth of "+21% to +25%" is an ex-Ampere-gain figure; the GAAP-comparable number is a 14–18% decline. `` Disclosed, correctly, in a footnote — and dropped by essentially all coverage.
Deferred revenue current rose $14,686M from $9,916M `` — consistent with the prepayment story rather than an independent flag.
Leverage and the rating. Total debt $125,337M (down $4.2B on senior-note repayment), equity $67,196M (up $24.1B, almost entirely the ATM). Gross debt/equity improved from ~3.0x to ~1.9x `` — but by issuing equity, not by generating cash. The S&P BBB− (2026-07-09, stable, OpenAI concentration named a "central credit risk") is one notch above speculative grade and was not re-checked this pass; whether the Q1 print or the equity raise moved any agency's view is open.
RPO cancellability: still not disclosable. The FY26 10-K contains no termination-for-convenience disclosure, no cancellation penalty, no minimum-take floor, and Oracle has elected the optional exemption not to disclose variable consideration allocated to wholly-unsatisfied obligations — meaning some usage-based consideration is excluded from the $664B while the fixed-vs-cancellable split of what is included is not broken out. `` The filings do not permit a conclusion either way. Anyone describing the $664B as "locked in" is asserting something the disclosure does not support; so is anyone describing it as soft.
Counterparty concentration: unchanged and un-closable. "No single customer accounted for 10% or more of our total revenues in fiscal 2026, 2025 or 2024" `` — a test on recognised revenue, which is close to meaningless when the backlog has barely begun converting. Oracle discloses no RPO concentration figure. This does not close with more reading; it closes via a counterparty's own filings or a named contract.
model.xlsx does not exist for this company and was not built. our figures holds five annual rows with revenue, margins and the entire balance-sheet block empty — only cash-flow items (FCF, OCF, capex, cash, shares, depreciation) are populated, auto-filled from SEC XBRL frames on 2026-09-04. Per the skill's own rule, generating a workbook seeded almost entirely from defaults would produce a valuation resting on invented balances. Opening balance sheet sourced: no. The projection below is prose ``, and the correct next action is to extract the balance-sheet block from the Q1 FY27 10-Q when it files — the Ex-99.1 balance sheet read above already supplies most of it and should be written to our figures in that pass.
Anchors, all primary:
, FCF −$23,686M, OCF $31,977M, capex $55,663M FY2027 (ending 2026-05-31 +1yr). Guidance is management's, is specific, and Q1 came in ahead of the run-rate it implies ($19.3B × 4 = $77.2B against ≥$90B, i.e. the guide requires acceleration through the year, which the Q2 cloud guide of +65–71% is consistent with). Base: revenue ~$90B, non-GAAP EPS $8.10 — taken as management's number, not re-derived. Free cash flow: capex $90–95B against an operating cash flow that, even at Q1's inflated $23.1B annualised, reaches ~$92B — FY27 FCF is negative on any reading, and negative by a wide margin ex-prepayments. The magnitude is not estimated here: it depends on the prepayment line, which has no disclosed forward path and which Oracle itself does not guide.
FY2028 and FY2029: NOT PROJECTED. This is the deliberate blank of this lens and it is the most important thing in it. The FY28–29 EPS path turns almost entirely on whether the months-13–36 tranche (~$108B/yr implied) converts on schedule, and on the financing-component prepayment line, which went from non-existent to half of operating cash flow in six quarters with no disclosed forward commitment, no counterparty named, and no guidance. A three-year EPS path built on top of those two unknowns would be a number that looks rigorous and is not — the exact failure mode the skill's provenance rules exist to prevent, and the exact reason the July dossier deferred the capex comparison until it could be done properly. It stays blank until the Q1 FY27 10-Q discloses the restated conversion split and the prepayment note.
No Brier forecast was logged. our model create is for a base case genuinely committed to; there is no committed base case beyond repeating management's own FY27 guide, and logging that as a forecast would score Oracle's guidance rather than this analysis.
The bull case, in its strongest form. Demand exceeds supply and customers are proving it with cash: $30B+ of new AI contracts booked in a quarter, 850MW and 300,000+ GPUs delivered (nearly triple Q4 FY26), IaaS +121%, and — the sharpest evidence — customers pre-paying $11.4B against future delivery. Customers do not lend money to a vendor whose capacity they could get elsewhere. The conversion curve is shortening at the margin for three straight quarters, so the backlog is moving toward the income statement rather than away from it. The equity raise was done into strength at a price the company chose. And beneath all of it sits a database annuity that has funded everything Oracle has ever done.
The bear case, in its strongest form — three ways this permanently impairs.
Pre-mortem — it is March 2028 and the thesis broke. The likeliest path, on what the filings show: the FY28 step-up did not arrive on time. The months-13–36 tranche converted slower than the curve implied because delivery, not demand, was the constraint — power, GPU supply and datacenter commissioning slipped, and RPO is a promise to deliver, not just to pay. Revenue grew fast and missed a guide anyway; the prepayment line, which was always a function of customers' own funding conditions rather than Oracle's, stopped growing at exactly the moment it was most needed; and Oracle returned to the equity market at a lower price. Note what is absent from that story: no fraud, no demand collapse, no lost customer. It is a timing failure on a balance sheet with no slack — which is why it is the plausible one.
What the market is refusing to see (contrarian read). Both sides. The bears are still telling a lengthening-backlog story that the last three 10-Qs contradict. The bulls are calling $23.1B of operating cash flow a record while 49% of it is borrowing from customers on which Oracle books interest expense. The two loudest claims about Oracle are each refuted by a disclosure sitting one line away from the number being quoted. That, rather than any directional view, is the defensible observation from this pass.
Beyond the July dossier's short case, this print adds three:
The record cash flow is a composition artifact and management chose to headline it anyway. "$23 billion, up 184%" is in the first paragraph of the release; the $11,363M prepayment line is on page four of the statements and was $0 in the comparative quarter. The reconciliation table Oracle itself provides — "Net Cash Outlay for Capital Expenditures" — exists precisely because management knows the gross number misleads. They built the bridge and led with the number on the wrong side of it.
They pre-answered the capital question nobody had asked yet. "the Company confirms there is no incremental impact on its plans to raise capital" is not a sentence that appears in a release unless funding is the question in the room. Oracle raised $19.9B of equity and $11.4B of customer credit in a quarter it spent $28.5B. The build is being funded by shareholders and customers, not by the business. That may be a perfectly rational way to finance a genuine land-grab — but it is the fact the coverage is not carrying.
The GAAP-comparable Q2 EPS guide is a decline. Down 14–18% including the prior-year Ampere gain. Presented as +21–25%. Disclosed in a footnote. Correct, and doing work.
What must hold for the current price: that the months-13–36 tranche converts roughly on schedule; that counterparty concentration does not realise; and that the prepayment channel either continues or is replaced without a DilutionIssuing new shares, so each existing share owns a smaller slice of the same company. raise at a worse price. All three are undisclosed or unguided.
the previous dossier, with the ordering changed by this print. The top three now are:(Question 1 was #1 in the June dossier and is still unanswered; it has now been unanswered across three dossiers and two prints.)
Lens 1 (business), Lens 2 (supply chain), Lens 3 (moats), Lens 4 (segments): unchanged — see the previous dossier. The Q1 FY27 revenue mix (cloud $11.6B, software $5.5B declining −3%, services $1.4B, hardware $0.8B) is consistent with the migration dynamic described there and moves no structural conclusion. Two product announcements in the release — the Oracle AI Data Platform (automated enterprise ontology generation, positioned explicitly against Palantir) and a 100% agentic health-care management/EHR system — are noted, not assessed: neither has disclosed revenue, pricing or customers.
Read it. Every Oracle figure here comes from Oracle's own SEC filings, read directly this session; the peer capex table comes from the peers' own XBRL facts. No position has ever been held in Oracle, no Oracle product has been operated, no operator has been spoken to, and the Q1 FY27 earnings call could not be reached. Per docs/brand/teaching-doctrine.md this is coverage standing, not teaching standing, and any published piece drawing on it must be labelled as such.
transcripts/ is empty for this company across every quarter. The 97.9% GPU utilisation and 20% renewal-premium claims remain unverified and unverifiable from here. Lens 6 is therefore unrun, not merely thin.n/a.our figures balance-sheet block is empty and model.xlsx does not exist. The Ex-99.1 balance sheet read in this pass supplies most of the block; write it in the 10-Q pass, then our model.regulatory/regulatory-findings.md is 85 days old — re-run our model next pass.Every dossier we have written on Oracle, newest first.
The Q1 FY27 print settles the two questions the July dossier could not
none (no verdict, no call this pass)
The FY26 10-K turns the bull narrative into audited fact and the bear narrative into a footnote you can now read
A 49-year-old database company has reinvented itself as the lender-of-last-resort GPU landlord
Covered in the Knowledge Base
Models
| Industry | Cloud Computing |
| Type | Tech Startup |
| Size | Public Company |
| Founded | 1977 |
| Website | Visit Oracle |
Where Oracle sits against the other names we cover on this beat. Each line is that company’s verdict, not a summary of it.
The 2026-07-30 print settled four of the five questions the prior dossier said it could
Cash $78.2B
The bear case arrived a year early and the bull case grew a new leg in the same quarter
Cash $90.3B
The buildout stopped being paid for by the ad business and started being paid for by the capital markets
Cash $90.3B
NOT ISSUED — the 2026-07-22 refresh withheld a directional call under the Socratic gate and this one keeps that discipline.
Cash $78.3B
NOT ASSESSED — this is a pre-print evidence refresh, not a call.