Oracle ORCL

Reference table

1. What Oracle now is — and the segment table the vendor does not have

Oracle reports three operating segments: cloud and software, hardware, and services. The company employs 141,000 people (49,000 in the US, 92,000 internationally) and spent $10.3 billion on research and development in FY2026.

The vendor payload has no FY2026 segment row at all — seg_prod and seg_geo both stop at FY2025, despite the 10-K having been filed 2026-06-22. We rebuilt FY2026 from the 10-K's Results of Operations deltas, and the reconstruction ties to reported revenue within $1 million:

Reported FY2026 revenue was $67.358 billion. The derivation ties.

Inside cloud, the mix shift is the whole story. The 10-K discloses that cloud infrastructure was 53% of total cloud revenues in FY2026, against 42% in FY2025 and 35% in FY2024 — and that in constant currency, cloud infrastructure contributed 84% of cloud revenue growth in FY2026 while cloud applications contributed 16%. Cloud applications, the higher-margin software half, are now the minority of cloud.

Geographically the growth is overwhelmingly American. The 10-K states that the Americas, EMEA and Asia Pacific contributed 88%, 5% and 7% respectively to constant-currency total revenue growth in FY2026. The vendor's last geographic split is FY2025: Americas $36.339B (63.3%), EMEA $14.025B (24.4%), Asia Pacific $7.035B (12.3%).

The margin consequence, from the payload's own income statements:

Gross margin fell 531 basis points in one year and 1,540 basis points in five. The 10-K corroborates the mechanism directly: "Total margin as a percentage of revenues in constant currency decreased in fiscal 2026 relative to fiscal 2025 due to an increase in total expenses driven by higher infrastructure expenses to support growth in our cloud infrastructure offering."

Operating margin held flat at 30.85% only because operating expenses grew slower than revenue — and because $1.8 billion of restructuring under the 2026 Restructuring Plan (total estimated cost up to $2.1 billion, driven in part by "the adoption and integration of AI technologies across certain functions") sat inside a line that also contains offsetting items. The margin structure of this business is changing fast, and gross margin is the honest place to watch it.

2. The single most important analytical issue: $638 billion of backlog against $260 billion of off-balance-sheet obligation

This is the crux of the name, and it requires three numbers held simultaneously.

First, the backlog is real, enormous and disclosed with a conversion schedule. From the FY2026 10-K:

> "As of May 31, 2026, our remaining performance obligations were $638 billion, of which we expect to recognize approximately 12% as revenues over the next twelve months, 34% over the subsequent month 13 to month 36, 34% over the subsequent month 37 to month 60 and the remainder thereafter."

The trajectory across the three filings and the prior-year comparatives is remarkable:

Remaining performance obligations grew 362% in one fiscal year, and were still adding $85.4 billion sequentially in Q4. Twelve per cent of $638 billion is $76.6 billion, which is 85.4% of the $89.614 billion FY2027 consensus revenue estimate. That is the strongest single fact in the bull case and it is verified in a filing six weeks old. Eighty per cent converts within five years — this is not a ten-year backlog dressed up as a two-year one.

Second, the cost of servicing it is committed and much of it is invisible on the balance sheet. From the 10-K's lease footnote:

> "As of May 31, 2026, we had $260 billion of additional lease commitments, substantially all related to data center arrangements, that are generally expected to commence between the first quarter of fiscal 2027 and fiscal 2029 and for terms of fifteen to nineteen years that were not reflected on our consolidated balance sheet... These additional lease commitments include a lease for which we have guaranteed up to $3.3 billion of the lessor's borrowing, which matures in September 2026."

And, in the same note:

> "Subsequent to May 31, 2026, we entered into an additional $19 billion of unconditional purchase commitments for cloud infrastructure assets that commence in fiscal 2027 and have a term of five years."

Stacking the obligations honestly:

Any leverage ratio computed from the vendor file understates Oracle's forward obligation by roughly a factor of three. We state the $260 billion as an undiscounted stream of future lease payments over fifteen-to-nineteen-year terms — it is not comparable to debt principal and we do not add it to enterprise value. But it is the reason the credit market is charging what it is charging, and it is entirely absent from every screen.

One nuance the bears should note: the off-balance-sheet commitment stopped growing in the fourth quarter. It was $248 billion at 2025-11-30, $261 billion at 2026-02-28 and $260 billion at 2026-05-31. Quarterly capital expenditure likewise stepped down: $20.5 billion in H1, $18.7 billion in Q3, $16.5 billion in Q4. That is either capital discipline or a demand pause, and the data cannot distinguish between them.

Third — and this is the fulcrum — the concentration of the backlog is not disclosed. The 10-K states:

> "No single customer accounted for 10% or more of our total revenues in fiscal 2026, 2025 or 2024."

That is a revenue disclosure. It says nothing about who holds the $638 billion of contracted future revenue. The same filing's risk factors are unusually direct about why that matters:

> "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... In addition, some of our customers may be highly leveraged and subject to their own operating and regulatory risks... In certain OCI offerings, we are more concentrated among a number of large customers, which could increase these risks."

Our knowledge base contains a specific, dated claim that roughly $500 billion of the backlog sits with a single very large counterparty (Section 7). We cannot confirm or refute it from the filings, and we will not pretend otherwise. The honest statement is: Oracle has disclosed a $638 billion asset and has not disclosed its counterparty concentration, while explicitly warning that concentration exists and that some counterparties are highly leveraged. That asymmetry — a fully disclosed obligation against a partially disclosed asset — is the reason this dive lands on Watch rather than Buy at a price that is otherwise defensible.

3. Growth — exponential revenue, arithmetic earnings

Consensus forward, on epsAvg only (the estimate block's EBITDA and net-income rows are rejected — Section 8):

The FY2027 row is the entire investment case in two numbers: revenue +33.0%, EPS +5.6%. A third of a year's revenue growth is consumed before it reaches the bottom line. That is not a modelling artefact; it is the arithmetic of a business that has swapped 80% gross margins for 65% ones, added $4.598 billion of interest expense and is about to add a great deal of depreciation.

The depreciation step, quantified. The 10-K discloses that the servers and networking equipment carry an estimated useful life of six years, and that depreciation on property, plant and equipment was $7.6 billion in FY2026, $3.9 billion in FY2025. Net PP&E closed FY2026 at $129.647 billion, up from $43.522 billion. Against an average base of $86.6 billion, $7.6 billion of depreciation is an effective rate of only 8.8% — an implied average life of about eleven years — because a large share of the base is construction in progress not yet in service (the 10-K's PP&E note describes one component as "servers, networking equipment and leasehold improvements to be deployed at our data centers"; the dollar split was not captured in the extracted filing text).

As that base is placed in service on six-year lives, the effective rate must rise toward 16.7%. On the FY2026-exit base of $129.647 billion alone, that is roughly $21.6 billion of annual depreciation — an increase of about $14 billion against FY2026 operating income of $20.778 billion. And capital expenditure is not slowing: the 10-K says "we expect this upward trend to continue during fiscal 2027 and in the following fiscal years."

This is the mechanism the market is pricing and the estimates are already partly reflecting. It is also precisely what one knowledge-base claim identified in April: the market "fears datacenter capex converts to depreciation and crushes profitability in 5-10 years" (Section 7).

4. Balance sheet — and the corrections

As at 2026-05-31, from the FY2026 10-K and the vendor balance sheet:

Test one: the double-counted-lease failure mode. Tested and PASSED. Short-term debt $7.199B plus long-term debt $122.342B equals $129.541 billion, which ties precisely to the 10-K's risk factor: "As of May 31, 2026, we had an aggregate of $129.5 billion of outstanding indebtedness that will mature between calendar year 2026 and calendar year 2066." The $26.648 billion of capitalised lease obligations is additive, not duplicative — a clean result on the failure mode that broke ROKU and TW. However, capitalLeaseObligationsCurrent is reported as zero, which cannot be right for a lease portfolio of that size, and the 10-K's lease-maturity table was not captured in the extracted filing text, so the current/non-current and operating/finance splits cannot be verified from these sources. Flagged rather than smoothed over.

Test two: net debt and enterprise value. The vendor is very nearly right — the only name in this batch where that is true. netDebt is reported at $124.900 billion, computed as total debt $156.189 billion less cash and equivalents $31.289 billion, omitting $605 million of short-term investments. Corrected net debt is $124.295 billion; corrected enterprise value is $419.742B + $124.295B = $544.037 billion against the vendor's $544.642 billion. The vendor overstates enterprise value by $605 million, or 0.11%. We report the correction for completeness and use the corrected figure, but no conclusion turns on it. The material enterprise-value problem here is not the vendor's arithmetic — it is the $260 billion of off-balance-sheet lease commitments that no enterprise-value calculation captures.

Test three: equity, and why every book-based ratio must be discarded.

• Total stockholders' equity $42.508 billion, of which $4.954 billion is the mandatory convertible preferred issued 2026-02-05. Common equity is therefore $37.554 billion, or $13.04 per share — not the vendor's bookValuePerShareTTM of $14.96, which includes the preferred and the $548 million minority interest. Price-to-common-book is 11.2x, not 9.87x.
• Tangible common equity is $42.508B − $62.261B goodwill − $3.229B intangibles − $4.954B preferred = approximately NEGATIVE $27.936 billion. The vendor's tangibleBookValuePerShareTTM of −$7.79 is directionally correct and confirms it.
• Retained earnings remain negative $4.309 billion after decades of buybacks, though they improved from −$15.481 billion on FY2026's earnings.
• Consequence: priceToBookRatioTTM (9.866), priceToFairValueTTM (9.866 — an identical figure and not a fair-value estimate at all), returnOnEquityTTM (50.6%) and financialLeverageRatioTTM (6.158) are all artefacts of a near-zero equity denominator and are REJECTED. A 50.6% return on equity at Oracle is not a quality signal; it is a leverage signal.

Test four: the capital structure Oracle actually built in FY2026. From the 10-K:

• $43.0 billion of senior notes issued, in thirteen tranches: 4.55% due Feb 2029; 4.45% due Sep 2030; 4.95% due Feb 2031; 4.80% due Sep 2032; 5.35% due May 2033; 5.20% due Sep 2035; 5.70% due Feb 2036; 5.875% due Sep 2045; 6.55% due Feb 2046; 5.95% due Sep 2055; 6.70% due Feb 2056; 6.10% due Sep 2065; 6.85% due Feb 2066. Net proceeds $42.7 billion.
• $5.0 billion of 6.50% Series D Mandatory Convertible Preferred Stock, issued 2026-02-05: 50,000 shares at a $100,000 liquidation preference, listed as depositary shares each representing 1/2,000th of a share. Converts automatically on 2029-01-15 into between 499.8126 and 624.7657 common shares per preferred share — implied conversion prices of $200.07 and $160.06. At today's $145.72 the stock is below the lower threshold, so holders would receive the maximum 31.24 million shares (1.08% dilution), worth roughly $4.55 billion against $5.0 billion of face. Dividends are $1,625 per share per quarter, $325 million a year.
• A $20 billion at-the-market common equity programme entered 2026-02-02, under which no shares had been sold as of 2026-05-31. At $145.72 that is 137.2 million shares, or 4.76% of the share count — a standing dilution overhang, and a clear statement of intent about how the next tranche of capital might be raised.
• The revolving credit agreement was increased from $6.0 billion to $10.0 billion in March 2026 and was undrawn at year-end; the commercial paper programme was increased to $10.0 billion with $1.5 billion outstanding.
• "We were in compliance with all debt-related covenants at May 31, 2026."
• Buyback authorisation remaining: $6.3 billion — but only $93 million (0.4 million shares) was repurchased in FY2026, against $600 million in FY2025 and $1.2 billion in FY2024. The buyback has been switched off.

A note on the fair value of Oracle's own debt, and against the simplest bear reading. The 10-K discloses that the $128.1 billion carrying value of senior notes and other long-term borrowings had an estimated fair value of $114.4 billion at 2026-05-31 — a 10.7% discount to carrying. That looks alarming until one reads the comparative: at 2025-05-31 the figures were $90.3 billion and $81.3 billion, a 10.0% discount. The discount is essentially unchanged year on year, which argues that most of it is duration on long-dated low-coupon paper rather than pure credit deterioration. We flag this because it cuts against our own bearish instinct and the reader deserves it. The cleaner credit signal is the new-issue curve: Oracle paid 4.55% for three-year money and 6.85% for forty-year money in the same fiscal year. That upward slope is the market charging more for Oracle's distant future than its near one — which is exactly the shape of a credit concern, and exactly what the knowledge base's credit lane describes.

5. Cash flow — the number that ends most arguments

In two years Oracle went from $11.8 billion of free cash flow to negative $23.7 billion. Operating cash flow grew 53.6% — an excellent result on its own — and was still not remotely enough. The 10-K confirms both figures: "$32.0 billion of cash inflows from our operations" and "Cash used for capital expenditures increased from $21.2 billion in fiscal 2025 to $55.7 billion in fiscal 2026 primarily due to the expansion of our data centers."

The cash bridge for FY2026, from the 10-K's liquidity discussion, is the clearest statement of what this company now is:

Sources: $42.7 billion net proceeds from senior notes · $5.0 billion from the mandatory convertible preferred · $4.9 billion from selling investments (principally a $4.3 billion equity-method disposal producing a $2.7 billion gain) · $3.3 billion of short-term financing related to capital expenditure · $1.3 billion net from employee stock programmes · $32.0 billion from operations.

Uses: $55.7 billion of capital expenditure · $5.8 billion of dividends to common and preferred holders · $5.0 billion of scheduled debt repayment · $851 million of net commercial-paper repayment · $864 million of investment purchases · $95 million of buybacks.

The dividend was borrowed. With free cash flow at negative $23.686 billion, every dollar of the $5.787 billion distribution came from the capital markets. That is a legitimate corporate choice during a buildout; it is not a sustainable one, and it is the reason the dividend — currently $2.00 a share, a 1.37% yield — belongs in the risk section rather than the return section.

Consequently the following vendor fields are arithmetically correct and analytically useless, and are REJECTED as valuation inputs: freeCashFlowYieldTTM −5.64%, priceToFreeCashFlowRatioTTM −17.72x, evToFreeCashFlowTTM −22.99x, freeCashFlowPerShareTTM −$8.23. The two that are useful, and that we do use, are capitalExpenditureCoverageRatioTTM at 0.574 (operating cash flow covers 57% of capital expenditure) and dividendPaidAndCapexCoverageRatioTTM at 0.520.

One more disclosure worth surfacing. The 10-K notes that "During fiscal 2026, we received $4.6 billion of prepayments from customers that included a significant financing component," against none in fiscal 2025 or 2024. Customers are now pre-funding Oracle's buildout. That is a positive demand signal and a negative signal about the cost of Oracle's alternatives, simultaneously.

6. Valuation — priced in or room?

At $145.72 (market cap $419.742B, ~2,880.5M shares, corrected enterprise value $544.037B):

Three adjustments that matter, all stated as ours rather than the vendor's:

(a) Reported FY2026 GAAP EPS of $5.83 is materially flattered. It contains a $2.7 billion realised gain on the disposal of an equity-method investment (closed 2025-11-25), and it was struck at an effective tax rate of 12.6% against the company's own adjusted rate of 19.9% — the difference arising from a legal-entity realignment benefit and a deferred-tax remeasurement under the US One, Big, Beautiful Bill Act. Stripping the gain from pretax income ($19.608B − $2.700B = $16.908B) and taxing at 19.9% gives $13.543 billion, or $4.61 a share after preferred dividends. The honest trailing GAAP multiple is 31.6x, not 25.0x.

(b) Consensus estimates are non-GAAP and exclude stock-based compensation of $4.811 billion — $1.65 a share, 20.5% of the FY2027 estimate. We do not treat stock compensation as a non-expense. On an SBC-inclusive basis FY2027 EPS is roughly $6.41 and the forward multiple is 22.7x, not 18.1x.

(c) EBITDA is a poor lens for this business and we barely use it. EBITDA of $32.138 billion excludes the depreciation that is now the dominant economic cost of the enterprise and the $4.598 billion of interest that funds it. The vendor's evToEBITDATTM of 16.97x is reported and not relied upon.

Peer context: there isn't any usable. The vendor peer set contains Helport AI ($19.5M market capitalisation), i3 Verticals ($522M), Radware ($1.24B) and Wix ($3.72B) alongside a $419.7 billion company. Only SAP, ASML, Adobe, Palantir, Synopsys and Fortinet are plausible in any sense, and none of them carries Oracle's leverage or capital intensity — which is precisely the point. No peer-multiple comparison is drawn.

The vendor's own composite model rates Oracle B− with an overall score of 2, and specifically scores debt-to-equity 1, price-to-book 1 and discounted cash flow 1 out of 5 — directionally consistent with everything above, and reported as corroboration rather than as an input.

6a. What today's price assumes (the inversion)

This is the interesting part on a name that has already halved. At $145.72 — 18.1x FY2027 consensus, 13.3x FY2028, and 9.4x FY2029 — the price is not paying for the growth at all. Reverse-engineered, it embeds roughly the following falsifiable claims:

• That FY2027 is close to the terminal earnings number. (Our arithmetic on consensus.) At a conventional 20x terminal multiple, $145.72 implies sustainable EPS of $7.29. Consensus FY2027 non-GAAP EPS is $8.058 — the market is paying less than a terminal multiple for a year the street thinks is 33% revenue growth. In plain terms: today's price assumes the FY2028 and FY2029 estimates ($10.93 and $15.45) do not happen.
• That the disclosed backlog conversion schedule is not achieved. (Filing-derived.) The 10-K schedules 12% of $638 billion — $76.6 billion — into the next twelve months, which is 85.4% of the $89.614 billion FY2027 consensus revenue estimate. A price at 18.1x forward on 33% contracted growth is only rational if that contracted revenue is at risk. This is the most falsifiable bullet in the dive: it is checkable every ninety days from the RPO note.
• That interest and depreciation absorb most of the incremental revenue. (Consensus-derived, and consensus already partly agrees.) Consensus models FY2027 revenue +33.0% and non-GAAP EPS +5.6%. Our own bound on the depreciation step — from $7.6 billion toward roughly $21.6 billion on the FY2026-exit property base at the disclosed six-year server life — plus interest expense scaling from $4.598 billion as borrowings grow past $129.5 billion, is consistent with that.
• That there is no credit event, and no forced equity issuance below current levels. (Our reading.) The $20 billion at-the-market programme is untouched but live; drawing it at $145.72 is 4.76% dilution. The knowledge base's credit lane reports credit-default-swap protection at 215bps and widening as of 2026-08-01. This is the most fragile assumption in the price, because unlike the others it is not a matter of degree — it is binary, and it is the one the equity market historically prices last.
• That the street's $249.18 consensus target is wrong by 41%. (Vendor-derived.) Fifty-six buys, twenty-six holds and four sells across 86 analysts, with a target range of $145 to $325 whose low end is exactly spot. Either the analyst community has not updated its models or the market has decisively rejected them. The gap between an unchanged $249 consensus and a $146 price is itself a datapoint: sell-side models are still running the FY2028-FY2029 conversion path, and the market is not.

6b. The return bridge (why the multiple moves)

Expected return ≈ EPS growth + multiple drift + shareholder yield − dilution.

• EPS growth is the weak leg near term and the strong leg later. Consensus non-GAAP EPS grows +5.6% in FY2027 (from $7.63 to $8.06) and +35.6% in FY2028. The near-term contribution to return is therefore almost nil; the case is a FY2028-FY2029 story, which is another way of saying the holding period is long and the intervening path is where the risk lives.
• Our base case assumes modest multiple EXPANSION, from 18.1x to 19.6x FY2027 consensus, and we flag that as the fragile leg. The justification is not optimism but normalisation: a business with 33% contracted forward revenue growth should not trade below a terminal multiple, and the current level implies active disbelief in disclosed backlog. But we deliberately keep the expansion small, because the offsetting force is real — as the depreciation base scales and free cash flow stays negative, the quality of the earnings deteriorates and a lower multiple can be perfectly rational.
• Shareholder yield is roughly zero and could be negative. The dividend is $2.00 a share, 1.37%, and it is currently funded by borrowing. The buyback is effectively dead ($93 million, 0.02% of market capitalisation). The $20 billion at-the-market programme is a potential −4.76%, and the mandatory convertible is a further −1.08% at prices below $160.06. Net shareholder yield is plausibly negative, which is unusual for a mega-cap and materially changes the return arithmetic.
• Putting it together: roughly +5.6% earnings growth, +8% from mild multiple normalisation, +1.4% dividend, less up to −5.8% of potential dilution, gives a central twelve-month expectation in the high single digits — which is exactly where our $158 base sits (+8.4%). Most of the return in the base case is neither growth nor re-rating; it is the absence of further de-rating. That is not a compelling risk-reward against a −41.7% bear case.

6c. Variant perception (where we differ, what would surprise)

A 44-45 knowledge-base split is not "no edge." It is a real disagreement, and this dive lands on a specific side of each of its two halves.

• We side with the BULLS on demand and with the BEARS on the financing. The backlog is not a narrative — it is a filed, audited, scheduled number that grew 362% in a year and was still adding $85 billion in the final quarter, with 80% converting within five years. The bulls who say "the demand is already there and contracted" are supported by the 10-K. The bears who say the buildout is a credit phenomenon are equally supported by the 10-K: 6.85% forty-year money, a 6.50% mandatory convertible, a $20 billion equity shelf, a dead buyback, a borrowed dividend and $260 billion of off-balance-sheet lease commitments. Both lanes are right about different halves of the same company, which is why the stock has halved and why it is now genuinely hard to value.
• We differ from the street's price target more than from its rating. A $249.18 consensus against a $145.72 price is a 71% gap, and the low target of $145 is spot. We think the sell-side models have not been re-based for the interest and depreciation step that consensus EPS itself already reflects — you cannot model +33% revenue with +5.6% EPS and simultaneously hold a target 71% above the market. Our $158 is 37% below consensus. Watchable event: whether targets are cut into or after the 2026-09-08 print.
• We differ from the simplest credit-panic read, in the bears' disfavour. The fair value of Oracle's long-term borrowings sits at a 10.7% discount to carrying — but it sat at a 10.0% discount a year earlier, before spreads widened. Most of that mark is duration, not distress. Oracle was covenant-compliant at year-end, has an undrawn $10 billion revolver, holds $31.9 billion of cash, and generated $32.0 billion of operating cash flow. This is a stretched balance sheet, not a broken one, and the distinction matters for position sizing.
• The concrete NEGATIVE surprise that would break the thesis: a sequential DECLINE in remaining performance obligations, or a fall in the twelve-month conversion percentage below 10%, at the 2026-09-08 print. Either would mean the backlog is lengthening or eroding rather than converting, against $260 billion of lease commitments that begin commencing in FY2027 regardless. Secondarily: any drawdown on the $20 billion at-the-market programme below $160, any rating downgrade (which per the 10-K could trigger collateral and letter-of-credit requirements and affect data-centre lease terms), or a disclosed impairment or renegotiation involving a large cloud counterparty.
• The concrete POSITIVE surprise that would force a re-rate: the RPO holding above $638 billion with the twelve-month conversion percentage rising above 12% — that combination is unambiguous, because it means the backlog is converting faster, which is exactly what the price says will not happen. Add capital expenditure guidance that peaks rather than continues rising, gross margin stabilising above 65%, credit-default-swap spreads back below 150bps, or any disclosure quantifying backlog concentration favourably. A restarted buyback against the $6.3 billion authorisation would be the loudest signal of all, because it would mean management believes the financing phase is over.

Synthos fair values

All three anchors are struck on consensus non-GAAP EPS (the only reliable row in the estimate block — Section 8), cross-checked on enterprise value to revenue and on the SBC-inclusive earnings basis.

• Bear ~$85 — 10.5x FY2027E ($8.058), or 13.3x the SBC-inclusive $6.41. Cross-check: enterprise value of $369.5 billion, 5.5x FY2026 revenue — note that this is not a distress multiple, which is deliberate: a genuine credit event takes the equity well below this. The scenario: a large counterparty inside the cloud backlog renegotiates or fails to perform; RPO conversion slips beyond the 12/34/34 schedule while $260 billion of lease commitments commence anyway; the at-the-market programme is drawn into weakness; spreads widen past 300bps and a downgrade triggers the collateral provisions the 10-K flags. −41.7%.
• Base ~$158 — 19.6x FY2027E, 14.5x FY2028E, 24.7x the SBC-inclusive $6.41. Cross-check: enterprise value of $579.6 billion, 6.5x FY2027E revenue and 4.4x FY2028E. The scenario: the backlog converts roughly on the disclosed schedule; FY2027 revenue lands near $89.6 billion of which $76.6 billion was already contracted; capital expenditure peaks across FY2027-FY2028; gross margin stabilises in the low-to-mid 60s; no equity is issued under the at-the-market programme; the multiple stops falling. +8.4%, plus a 1.37% dividend that is currently borrowed.
• Bull ~$265 — 24.3x FY2028E ($10.925) or 17.2x FY2029E ($15.446). Cross-check: enterprise value of $887.6 billion, 6.8x FY2028E revenue. The scenario: conversion runs on or ahead of schedule to $130 billion of FY2028 revenue; the depreciation step is absorbed because the incremental contracts were priced to carry it; free cash flow crosses positive in FY2029; spreads compress and the equity re-rates from "levered AI proxy" to "contracted-backlog infrastructure compounder." +81.9%. Still 6% below the street's current $283 average of high and consensus.

Base is 8.4% above spot; the raw asymmetry is roughly 2:1 to the upside (−42% down, +82% up). But that ratio flatters the case, because the bear scenario is credit-shaped and the bull scenario is schedule-shaped — the downside can arrive in a fortnight and the upside takes three years. With base fair value inside 10% of spot, a genuinely split conviction lane, an undisclosed concentration at the centre of the thesis and a print in 35 days, this is a Watch.

7. Knowledge base — 103 claims, 44 bullish, 45 bearish, the most contested lane in this batch

Raw hits naming Oracle: 104. Searches covered ORCL and Oracle across all 51,928 records. One record was discarded: a quarantine_misattributed claim from 2026-05-23 (bullish, conviction 55, on the "SaaS apocalypse" narrative), excluded under the attribution policy. Usable: 103 claims across 24 independent sources, spanning 2018-01-16 to 2026-08-01 — the lane runs to three days before this dive.

Weighting applied, and stated:

• Management-class (HALF weight), 4 claims: jensen_huang (2), sam_altman_mgmt (1), andy_jassy_mgmt (1). These are chief executives of other companies discussing their own commercial interests; none is Oracle's own management, but all are promoting businesses whose economics depend on the buildout continuing.
• Promoter-class (LOW weight), 1 claim: anthony_pompliano in his own voice (2026-08-01, bullish, conviction 72). Per house rule this carries low weight.
• anthony_pompliano_show (9 claims) is treated as first-class GUEST lane. The claim texts read as guest positions in every case — "the most misguided call of his career," "he believes in AI's future but says AI stocks are now more overvalued than the internet was in 2000" — and are attributed in the third person. We cannot verify per-claim speaker identity from the stored records, and we say so rather than assuming. Notably these nine claims split 3 bullish / 6 bearish, so the treatment does not tilt the conclusion.
• Independent-voice count: 98 claims across 20 independent sources after removing the quarantined record, the four management-class claims and the one promoter-class claim.

The lane is not undecided. It is sequenced, and the sequence is the analysis.

Phase one — the backlog shock (June to October 2025), overwhelmingly bullish:

> 2025-09-13 · jordi_visser · bullish · conviction 90

> "Oracle's $300B+ quarterly backlog growth signals AI demand outstripping supply; scale of moves proves AI is the dominant force."

> 2025-09-14 · jordi_visser · bullish · conviction 82

> "Oracle's RPO orders up 359% YoY to ~$455B, $317B added QoQ; inference demand insatiable, revenue already booked — the inference buildout is the next phase and it's real, not a bubble."

This claim can be checked against the filings, and it checks out. The FY2026 10-K reports RPO of $138 billion at 2025-05-31 and the Q2 10-Q reports $523.3 billion at 2025-11-30; a Q1 figure near $455 billion is entirely consistent. A knowledge-base claim that reconciles to an SEC filing to within a few per cent is worth more than one that does not, and this one does.

> 2025-06-20 · bill_gurley · bullish · conviction 65 — "Oracle, left for dead in the 2000s, SaaS and AI eras, is reinventing itself—jumping from 5% cloud revenue share to 19% of Nvidia GPU allocation."

> 2025-11-26 · forward_guidance · bullish · conviction 75 — "Every cloud provider faces capacity constraints; Oracle stock is being hit because it can't meet the compute revenue demand arriving."

Phase two — the credit turn (November 2025 to March 2026), overwhelmingly bearish. This is the lane the brief asked us to surface in full, and it is load-bearing.

> 2026-08-01 · eurodollar_university · bearish · conviction 55 · thesis

> "AI-buildout debt is drawing growing caution — CDS on Oracle (215bps), SpaceX, Nvidia, Meta, Amazon, Alphabet moved higher in July; Amazon's $25B deal saw its order book shrink from $62B to $41B."

> 2026-07-30 · eurodollar_university · bearish · conviction 72 · thesis

> "The AI bubble is a credit bubble, not an equity one — CDS spreads and weak bond demand show the debt cycle turning as capex is financed at the wrong price, risking terrible returns."

These two are the freshest and most specific bear claims in the lane, and the filings corroborate the mechanism rather than the level. We cannot independently verify a 215bps credit-default-swap quote — it is not in any filing or in the vendor payload, and we do not have market credit data. What we can verify is everything that would produce it: Oracle paid 6.85% for forty-year money and 6.70% for thirty-year money in February 2026, issued a 6.50% mandatory convertible preferred rather than straight debt, registered a $20 billion at-the-market equity programme, stopped its buyback, and carries $260 billion of off-balance-sheet lease commitments against $129.5 billion of borrowings. A 215bps spread on a nominally investment-grade software company is an extraordinary datapoint, and the FY2026 10-K is entirely consistent with it. An independent corroboration from six months earlier:

> 2026-02-09 · a16z · bearish · conviction 40 — "Oracle's bet-the-company move into becoming a cloud commits very large capital, will go cash-flow negative for many years, and its credit default swap cost has risen to ~2% over three months — a risk he's monitoring."

Two hundred basis points in February, 215 in August. Two independent sources, six months apart, converging on the same number.

The rest of the credit lane, in date order:

> 2025-11-14 · forward_guidance · bearish · conviction 70 — "Oracle CDS blowing out as it levers up on debt to fund AI buildout; arbitrageurs short the equity as spreads widen, causing the stock to roll over."

> 2025-11-19 · luke_gromen · bearish · conviction 70 — "Hyperscalers shifted from cash-flow to creative debt financing; Oracle credit spreads and CDS are rising sharply, an early stress signal."

> 2025-11-28 · compound_and_friends · bearish · conviction 70 — "Oracle most egregious on capex—expected to spend 52% of revenue (up from 10% in 2021); $1.4T of obligations over 8 years is alarming."

> 2025-11-28 · compound_and_friends · bearish · conviction 55 — "Roughly $500B of Oracle's RPO obligations tied to [a single very large AI-infrastructure counterparty]; concentration of counterparty risk is underappreciated."

> 2025-12-18 · arthur_hayes · bearish · conviction 55 — "Oracle is the exception to bullish AI — increasingly debt-fueled rather than spending CapEx out of cash."

> 2025-12-21 · jordi_visser · bearish · conviction 70 — "Free-cash-flow negative with heavy debt and no AI revenues yet; significant problem if revenues don't arrive by 2028."

> 2026-03-30 · anthony_pompliano_show · bearish · conviction 72 — "Oracle turned from $30B free-cash-flow positive to roughly $10-15B negative as it burns cash chasing AI buildout."

Three of these can be scored against the filing, and the scoring is instructive:

1. The 52%-of-revenue capital expenditure claim UNDERSTATED reality. Actual FY2026 capital expenditure was 82.6% of revenue ($55.663B / $67.358B). The bear was not bearish enough.

2. The "$10-15B negative free cash flow" claim also UNDERSTATED reality. Actual FY2026 free cash flow was negative $23.686 billion.

3. The concentration claim cannot be verified. The 10-K discloses only that no single customer exceeded 10% of revenue; it discloses nothing about backlog concentration. We neither endorse nor dismiss the ~$500 billion figure, and the fact that Oracle has not disclosed it — while its own risk factors concede "in certain OCI offerings, we are more concentrated among a number of large customers" and "some of our customers may be highly leveraged" — is itself the finding. (Counterparty names in this claim are redacted per house style; the substance is preserved.)

Phase three — the two-sided fight (February to August 2026). This is where the 44-45 split actually lives.

The bull side, on valuation and oversold conditions:

> 2026-02-06 · compound_and_friends · bullish · conviction 65 — "The AI winners are in an 'anti-bubble': Oracle down 55%, Microsoft down 26% — nobody wants to own them, suggesting they're oversold rather than overvalued."

> 2026-05-21 · raoul_pal · bullish · conviction 65 — "Everyone's going to be wrong on Oracle; it probably goes back to new all-time highs."

> 2026-07-03 · compound_and_friends · bullish · conviction 70 — "Sticking with Oracle—an early AI adopter and former value tech name being unjustly punished on AI fears despite the [large AI counterparty] relationship."

> 2026-07-25 · jordi_visser · bullish · conviction 74 — "Not worried about hyperscaler debt — low debt/equity, ~$2T of RPOs and ~$2T of compute already contracted across Amazon, Microsoft, Google and Oracle; demand is already there and CapEx keeps accelerating."

Note the specific weakness in that last claim: "low debt/equity." Oracle's debt-to-equity is 3.67x and its tangible common equity is negative $27.9 billion. The generalisation across four companies does not survive contact with Oracle's individual balance sheet, and this is the clearest case in the lane where a bullish claim is contradicted by the filing. We downweight it accordingly, and note that jordi_visser is simultaneously the single largest contributor to the bearish side of this lane (his own claims across 2025-11 to 2026-03 include "if picking the single highest-risk name it's Oracle").

The bear side, on structure rather than credit:

> 2026-03-31 · compound_and_friends · bearish · conviction 65 — "Oracle down 58% has clean risk/reward but isn't a fat pitch; it never belonged at those highs and [the large AI counterparty] can't fund its committed spend."

> 2026-04-10 · compound_and_friends · neutral · conviction 68 — "Mag 7 valuations compress despite record earnings — market fears datacenter capex converts to depreciation and crushes profitability in 5-10 years."

> 2026-07-13 · compound_and_friends · bearish · conviction 55 — "Spender stocks (Meta, Oracle) no longer reacting positively to capex is a 1999-style early warning."

The 2026-04-10 claim is the most analytically useful in the entire lane, because it names the exact mechanism we quantified independently in Section 3 from the six-year server life and the $129.647 billion property base. A claim that identifies the mechanism before the numbers confirm it is worth more than one that reacts to the price.

And the counterweight to the credit lane, in Oracle's specific defence:

> 2026-02-13 · forward_guidance · neutral · conviction 60 · principle — "Hyperscalers have huge runway to fund capex by tapping cheap debt (Google 100-yr bonds) plus equity (Oracle) — no bankruptcy concern, but rotation can still de-rate them."

That claim, from February, described the $20 billion at-the-market programme and the mandatory convertible before the 10-K confirmed them, and it draws the correct distinction: solvency risk and de-rating risk are different risks. We agree with it, and it is the single sentence closest to our own conclusion.

> 2025-12-15 · andreas_steno · bullish · conviction 60 — "Oracle's negative free cash flow vs positive net income is just AI capex, normal for a growth stock; not an issue unless the AI investments fail to return."

Why "Mixed" and not a directional call. Twenty independent sources, 98 independent claims, 44 bullish against 45 bearish, running to three days before this dive. We have not seen a lane this evenly balanced on a mega-cap. But the balance is not symmetric in quality: the bull lane is strongest on demand and weakest on balance sheet (the "low debt/equity" claim is simply wrong for Oracle); the bear lane is strongest on financing and weakest on demand (it has consistently underestimated the backlog). Our verdict follows that shape exactly: we take the bulls' demand case and the bears' financing case, and the result is a Watch.

8. Data integrity — what we rejected from the vendor file and why

Fifteen items. Two are filing-versus-vendor conflicts, both named.

1. FILING VS VENDOR — buyback. cf_a.commonStockRepurchased reports −$206M (FY2026) and −$1.5B (FY2025). The 10-K reports $93 million (0.4 million shares) and $600 million (3.9 million shares) under the repurchase programme, and the MD&A separately states "$95 million of cash used for repurchases of our common stock" in FY2026. The vendor overstates the FY2026 buyback by 2.2x and the FY2025 buyback by 2.5x. THE FILING WINS: we use $93 million and $600 million. The likely reconciling item is share-withholding for employee tax obligations, which is not a repurchase-programme buyback. The conclusion — the buyback has been switched off — depends on the filing figure, not the vendor's.

2. FILING VS VENDOR — segment data. seg_prod and seg_geo have NO FY2026 rows, stopping at FY2025, despite the FY2026 10-K having been filed 2026-06-22. THE FILING WINS: we rebuilt FY2026 segments from the 10-K's Results of Operations deltas (cloud and software +$9.3B → ~$58.530B; hardware +$148M → ~$3.084B; services +$510M → ~$5.743B), a reconstruction that ties to reported revenue of $67.358 billion within $1 million. Geographic FY2026 revenue is likewise unavailable; the 10-K's growth-contribution percentages (Americas 88%, EMEA 5%, Asia Pacific 7%) are used instead and labelled as such.

3. est net-income and EBITDA rows are on a NON-GAAP basis and are not comparable to GAAP actuals — REJECTED. FY2026E netIncomeAvg of $21.328B against actual GAAP net income of $17.087B (+24.8%); FY2026E ebitdaAvg of $26.792B against actual EBITDA of $32.138B (−16.6%). This is the estimate-block corruption the data contract warns about. epsAvg was verified against the earnings calendar (FY2026 non-GAAP actuals sum to $7.63 against a $7.487 estimate — a 1.9% variance) and is the only estimate row used anywhere in this dive.

4. capitalLeaseObligationsCurrent is reported as ZERO — implausible and unverifiable. A $26.648 billion lease portfolio must have a current portion. The 10-K's lease-maturity table was not captured in the extracted filing text, so the current/non-current split and the operating-versus-finance split cannot be reconstructed. Flagged rather than estimated around.

5. The lease double-count failure mode was TESTED AND PASSED — reported because a negative result is informative. shortTermDebt ($7.199B) + longTermDebt ($122.342B) = $129.541 billion, which ties to the 10-K's "$129.5 billion of outstanding indebtedness." capitalLeaseObligations of $26.648 billion is additive, not duplicative. Unlike ROKU and TW, Oracle's vendor debt figures are clean.

6. netDebt omits $605M of short-term investments — a 0.5% error, the smallest we have found. Reported $124.900 billion; corrected $124.295 billion. Enterprise value corrected from $544.642 billion to $544.037 billion, an overstatement of 0.11%. Documented for completeness; no conclusion turns on it.

7. THE MATERIAL BALANCE-SHEET GAP IS NOT A VENDOR ERROR — it is that $260 billion of lease commitments are off balance sheet by construction and appear in no field of the payload. Nor do the $19 billion of post-year-end purchase commitments or the $3.3 billion lessor guarantee. Every leverage ratio computable from this file understates Oracle's forward obligation by roughly a factor of three. This is the single most important thing the filings added to this name.

8. All book-value-based fields REJECTED. bookValuePerShareTTM ($14.96) includes the $4.954 billion mandatory convertible preferred and $548 million of minority interest; common book value per share is $13.04. priceToBookRatioTTM (9.866) should be 11.2x. tangibleBookValuePerShareTTM (−$7.79) is directionally right — tangible COMMON equity is approximately negative $27.936 billion. priceToFairValueTTM (9.866) is numerically identical to price-to-book and is not a fair-value estimate; rejected outright.

9. returnOnEquityTTM of 50.6% REJECTED as a quality signal. It is an artefact of a $42.5 billion equity base supporting $261.8 billion of assets after decades of buybacks. financialLeverageRatioTTM of 6.158 is the same fact stated honestly. The useful return metric is returnOnInvestedCapitalTTM at 7.99%, which we do use — and which sits at or below the 6.5%-6.85% marginal cost of Oracle's new capital.

10. All free-cash-flow-based multiples REJECTED as valuation inputs. freeCashFlowYieldTTM (−5.64%), priceToFreeCashFlowRatioTTM (−17.72x), evToFreeCashFlowTTM (−22.99x), freeCashFlowPerShareTTM (−$8.23) are arithmetically correct on free cash flow of −$23.686 billion and analytically meaningless. We use capitalExpenditureCoverageRatioTTM (0.574) and dividendPaidAndCapexCoverageRatioTTM (0.520) instead, which state the same fact usefully.

11. Trailing GAAP earnings are FLATTERED and the reported P/E is understated — adjusted by us. FY2026 GAAP diluted EPS of $5.83 contains a $2.7 billion realised gain on an equity-method disposal and a 12.6% effective tax rate against the company's own adjusted 19.9%. Adjusted, the figure is $4.61 and the trailing multiple is 31.6x, not the vendor's priceToEarningsDilutedRatioTTM of 24.99x. The vendor's number is arithmetically right and economically misleading.

12. Internal conflict WITHIN the vendor file on the 52-week range and moving averages. quote.yearHigh $345.72 / yearLow $114.50 against tech.hi52 $328.33 / lo52 $114.99; quote.priceAvg50 $164.92 against tech.sma50 $163.11; priceAvg200 $183.12 against sma200 $181.55. The tech block is computed on closing prices from six years of daily data; the quote block appears to use intraday extremes. We use quote for the range (giving −57.9% from the high) and tech for RSI, MACD and relative returns, and say so. On the tech close basis the drawdown is −55.6%.

13. peers is unusable. Helport AI ($19.5M), i3 Verticals ($522M), Radware ($1.24B) and Wix ($3.72B) against a $419.7 billion company. No peer-multiple comparison is drawn anywhere in this dive.

14. profile.ceo reports a single chief executive (Michael D. Sicilia). The 10-K's legal-proceedings note refers to "our two Chief Executive Officers" and separately to an Executive Chair and Chief Technology Officer; the executive-officer section identifies Safra Catz as Executive Vice Chair since September 2025 and Hilary Maxson as Chief Financial Officer since April 2026. The vendor's governance field is incomplete for a name where management change is itself material.

15. Earnings-call transcripts are NOT available on this plan. There is no management commentary on backlog conversion, counterparty quality, capital-expenditure trajectory or financing plans beyond what the filings state. On a name where the central question is undisclosed counterparty concentration, this is the most consequential gap in the dataset, and it is a reason the verdict is Watch rather than a directional call. 8-K exhibits carry the earnings release, not the call.

Non-equity tripwire: NOT triggered for this security, but note the adjacent one. ORCL at $145.72 with beta 1.712, a $114.50–$345.72 annual band and 36.5 million shares of volume is unambiguously common equity. However, Oracle also has a listed preferred — depositary shares each representing a 1/2,000th interest in the 6.50% Series D Mandatory Convertible — which is a par-like, fixed-dividend instrument that would trip the tripwire if it appeared in this feed. It does not. Any future dive on that security must not value it as common equity.

The three 8-Ks, identified. The brief anticipated a debt-issuance 8-K; there is none in this set. The three are: 2026-06-10 (Q4 FY2026 results under Item 2.02, plus declaration of a $1,625-per-share preferred dividend and a $0.50 common dividend); 2026-05-12 (election of Tomislav Mihaljevic to the board); 2026-04-06 (appointment of Hilary Maxson as Chief Financial Officer effective the same day, with a $950,000 salary, a $2.5 million target bonus and a $26 million equity grant whose performance tranche vests against "certain revenue metrics"; Douglas Kehring stepped down as Principal Financial Officer). The $43.0 billion of FY2026 senior-note issuance is documented in the 10-K, not in any 8-K in this set.

9. Technicals — a 27% bounce off the low, not an oversold entry

• Price $145.72, +2.73% today on 36.5 million shares against a 30.3 million average — a 1.20x volume day, meaningful but not a capitulation or a breakout.
• −57.9% from the 52-week high of $345.72; +27.3% above the 52-week low of $114.50. On the closing-price basis in the tech block the drawdown is −55.6% from $328.33 and the low is $114.99.
• Below BOTH moving averages, both declining: −11.6% under a 50-DMA of $164.92 and −20.4% under a 200-DMA of $183.12. MACD is −6.55.
• RSI is 59.7. This is the most important technical fact in the section and it cuts against the contrarian reading: the stock is not oversold. It has already run 27% off the low and is approaching neutral-to-overbought on a fourteen-day basis while still trading 12% below a falling 50-day average. That is the classic shape of a bear-market rally into resistance.
• Relative performance is the worst in this batch by a wide margin: 12-month −40.4% against SPY +24.3% and QQQ +30.8% — deficits of 65 and 71 percentage points; 6-month −8.9% against SPY +11.1%; 3-month −19.2% against SPY +7.6%. The three-month figure is the telling one: the underperformance is recent and ongoing, not a stale wound.
• Sentiment is NOT at a trough, and that matters. Fifty-six buy ratings against twenty-six holds and four sells, with a consensus target of $249.18 — 71% above spot and a low target of $145 that is essentially spot. This is not a capitulated name; it is a name whose covering analysts have not yet capitulated. That is a very different setup from a genuine sentiment bottom, and it is a reason to expect further target cuts rather than upgrades into the 2026-09-08 print.
• What today's +2.73% does to entry: it worsens it. Buying a 2.7% up day, at RSI 60, 12% below a declining 50-day average, 35 days before a print delivered by a four-month-old chief financial officer, is paying for a bounce rather than accumulating into weakness. If the thesis is right, the price will be available again; if it is wrong, the $114.50 low is the reference point, not the entry.

10. Insiders — one seller, and a large one

The reading, stated carefully. The Vice Chairman exercised 400,000 options at $40.93 and sold 327,137 shares — 81.8% of the exercised block — the same day, across six tranches at $156.06 to $160.86. Proceeds were roughly $51.6 million. The position went from 400,000 shares to 72,863.

Three qualifications, because exercise-and-sell is common and should not be over-read. First, the options were struck at $40.93 and had presumably neared expiry — a same-day exercise-and-sell is the standard mechanic and often reflects nothing about conviction. Second, this is a single individual, not a pattern across the executive team. Third, the only other transaction in the file is a routine restricted-stock award to the Chief Legal Officer at a $0 price.

What it does establish is the absence of the opposite. The transactions occurred fourteen days after the FY2026 results, at prices 7% to 10% above where the stock trades today, and not one insider bought a single share on the open market across a period in which the stock traded between roughly $115 and $180. In a name where the equity has halved and management is asking the capital markets to fund a $55 billion annual buildout, the absence of any insider buying is a mild negative, and the scale and timing of the one sale make it a slightly firmer one.

Note for context: the vendor file contains only the last eight insider transactions, so this is not a complete twelve-month picture and should not be read as one.

11. Verdict, kill-criteria and flip conditions

Watch.

The question the brief posed — broken thesis or violent de-rating of an intact one — has a specific answer: the thesis is intact and the company is different. Oracle's demand is not in doubt; $638 billion of contracted remaining performance obligations, up 362% in a year and still adding $85 billion in the final quarter, with 80% converting inside five years, is about as hard as forward evidence gets. What changed is what has to be true for that demand to reach the equity holder. In FY2024 Oracle turned $52.961 billion of revenue into $11.807 billion of free cash flow. In FY2026 it turned $67.358 billion of revenue into negative $23.686 billion, spent 82.6% of revenue on capital expenditure, borrowed its dividend, stopped its buyback, issued 6.85% forty-year paper and a 6.50% mandatory convertible, registered a $20 billion equity shelf, and committed to $260 billion of lease payments that are not on its balance sheet. The 57.9% de-rating from $345.72 was not an overreaction; it was a repricing of a genuinely different security.

At $145.72 the price has stopped paying for the growth. Eighteen times FY2027 consensus, thirteen times FY2028 and nine times FY2029 on a business the street models at +33%, +45% and +38% revenue growth — with 85% of the FY2027 revenue number already contracted — is a price that says the schedule will not be met. That is a real, falsifiable disagreement with the filing, and it is why this is not an Avoid.

It is not a Buy for one reason above all others: the single fact that decides the case is not disclosed. Oracle has published a $638 billion asset and has published nothing about its counterparty concentration, while writing in its own risk factors that "some of our customers may be highly leveraged" and that "in certain OCI offerings, we are more concentrated among a number of large customers." Our knowledge base contains a specific claim that roughly $500 billion of it sits with one counterparty. We cannot verify it and we will not pretend to. A $260 billion off-balance-sheet lease obligation against a backlog of undisclosed concentration is an asymmetry a Buy rating cannot survive. Add a base fair value only 8.4% above spot, a knowledge-base lane split 44-45 with the credit side more recent, a stock that has already bounced 27% off its low to RSI 60, and a print 35 days away under a four-month-old chief financial officer — and Watch is the honest verdict.

Pre-registered KILL criteria — what would move this to Avoid:

• Remaining performance obligations DECLINING sequentially, or the disclosed twelve-month conversion percentage falling below 10% (it was 10% at 2025-11-30, 12% at 2026-02-28 and 12% at 2026-05-31). A rising balance with a falling conversion rate is backlog lengthening, not demand strengthening — and the lease commitments commence regardless.
• Any drawdown on the $20 billion at-the-market equity programme below ~$160. Issuing 4.8% of the company at a 54% discount to the prior-year high would be an admission that the debt markets have closed, and it is the most likely path to a lower low.
• A credit-rating downgrade. The 10-K is explicit that this could "affect the terms or availability of certain long-term commitments (including data center leases)... and increase collateral, letter of credit or other credit support requirements under certain contractual arrangements." A downgrade is not a headline here; it is a cash call.
• Credit-default-swap protection widening decisively beyond 300bps, or a failed or heavily concessionary senior-note issuance.
• Any disclosed impairment, renegotiation, restructuring or non-payment involving a large cloud counterparty, or a write-down against the property base (the 10-K notes no significant property impairment in FY2026, FY2025 or FY2024 — that streak breaking would be material).
• Gross margin falling below 60% on a fiscal-year basis (65.20% in FY2026, 70.51% in FY2025), which would mean the incremental infrastructure revenue is being taken at a price that does not carry its capital cost.
• A dividend cut. It would be the correct capital-allocation decision and a devastating signal, and both facts should be held at once.

Pre-registered FLIP TO BUY (and upsize) conditions:

• The 2026-09-08 print showing RPO at or above $638 billion WITH the twelve-month conversion percentage at or above 12%. This is the single cleanest flip trigger in the dive: it directly falsifies the central claim embedded in today's price.
• Operating cash flow of $9 billion or more in Q1 FY2027 (against $31.977 billion for FY2026) with capital-expenditure guidance that peaks rather than continuing to rise.
• Any disclosure quantifying backlog concentration favourably — for example a stated maximum single-counterparty share of remaining performance obligations below 30%. This alone would move the verdict.
• A restarted buyback against the $6.3 billion remaining authorisation, which would signal management believes the financing phase is behind it.
• Credit-default-swap protection back below 150bps, or a senior-note issuance at coupons meaningfully inside February 2026's 6.70%-6.85% long end.
• A price below ~$110 — through the 52-week low — without any of the kill criteria having triggered. At that level the FY2027 multiple is 13.7x and the FY2028 multiple 10.1x on contracted revenue, and the risk-reward becomes asymmetric enough to act on despite the undisclosed concentration.

Where ORCL fits in the Synthos Framework Portfolio. No position today. If the flip conditions trigger, the AI-infrastructure / levered-growth sleeve at a 1-1.5% starter weight — deliberately below the batch norm, because the bear case is credit-shaped and credit outcomes are not gradual. This is explicitly not a name to build over weeks the way a compounder is built; it is a name to buy at a defined trigger or not at all. Today's +2.73% bounce, at RSI 60 and 11.6% below a declining 50-day average, is the wrong entry regardless of the eventual verdict. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $145.72.

Single biggest risk: undisclosed counterparty concentration inside the $638 billion backlog. Every other risk in this dive is sized, dated and traceable to a filing. This one is the only material fact that Oracle has chosen not to disclose, and it happens to be the one that determines whether $260 billion of commencing lease obligations are matched by a payer. The 10-K confirms no single customer exceeded 10% of revenue and warns in the same breath that "some of our customers may be highly leveraged." A concentration disclosure — in either direction — would move this dive's verdict more than any other single datapoint.

Most fragile assumption in the price: that the FY2027 consensus is roughly the terminal earnings number. At $145.72, a 20x terminal multiple implies sustainable EPS of $7.29 against a FY2027 consensus of $8.058 — the market is paying a discount to a terminal multiple for a year that is 85% contracted and modelled at +33% revenue. If the FY2028 and FY2029 estimates are even half right, the stock is materially cheap. If the schedule slips or a counterparty fails, the estimates collapse together and there is no tangible book, no free cash flow and no buyback to floor the price. That is the entire trade in one sentence, and it is not resolvable from the data currently disclosed.

Provenance & disclosures

• Traceability: 104 raw knowledge-base hits name Oracle; 1 was discarded (a quarantine_misattributed record dated 2026-05-23, excluded under the attribution policy), leaving 103 usable claims across 24 independent sources spanning 2018-01-16 to 2026-08-01. Searches covered ORCL and Oracle across all 51,928 records. Stance split: 44 bullish / 45 bearish / 14 neutral — net conviction MIXED, the most contested lane in this batch. Principal sources: jordi_visser (30 claims, himself split between a bullish demand lane and a bearish balance-sheet lane), compound_and_friends (23), forward_guidance (10), anthony_pompliano_show (9, treated as first-class guest lane — see below), all_in (4), eurodollar_university (2, the freshest and most specific bear claims). Weighting applied and stated in Section 7: 4 claims are management-class at HALF weight (jensen_huang ×2, sam_altman_mgmt, andy_jassy_mgmt); 1 claim is promoter-class at LOW weight (anthony_pompliano in his own voice, 2026-08-01). Independent-voice count after those exclusions: 98 claims across 20 independent sources. The 9 anthony_pompliano_show claims are treated as first-class guest positions because the stored text is third-person and reads as guest commentary in every case — we cannot verify per-claim speaker identity from the stored records and say so explicitly; those 9 split 3 bullish / 6 bearish, so the treatment does not tilt the conclusion. All quotes are verbatim from the stored claim text, except that names of specific AI models, laboratories and model families have been redacted and replaced with bracketed generic descriptions per house style; no substance was altered.
• Data as-of: fundamentals — FY2026 annual statements to 2026-05-31, from the 10-K filed 2026-06-22, six weeks before this dive; quarterly income statement to 2026-05-31 (Q4 FY2026); no quarterly balance sheet or cash-flow statement is present in the payload · estimates 2026-08-04 · prices 2026-08-04 ($145.72; 50-DMA $164.92; 200-DMA $183.12; RSI 59.7; MACD −6.55) · knowledge-base claims 2026-08-04. Filings used: 10-K 2026-06-22 (FY2026, primary source), 10-Q 2026-03-11 (Q3 FY2026), 10-Q 2025-12-11 (Q2 FY2026), 8-K 2026-06-10, 8-K 2026-05-12, 8-K 2026-04-06. No figure comes from memory, recall or external retrieval.
• Filing-versus-vendor conflicts (Section 8, items 1 and 2), stated explicitly: (i) the vendor reports FY2026 share repurchases of $206 million and FY2025 of $1.5 billion; the 10-K reports $93 million (0.4 million shares) and $600 million (3.9 million shares) under the repurchase programme, and the MD&A states "$95 million of cash used for repurchases" — we use the filing, and the conclusion that the buyback has been switched off depends on it. (ii) the vendor's seg_prod and seg_geo have no FY2026 rows at all; we rebuilt FY2026 segment revenue from the 10-K's Results of Operations deltas, tying to reported revenue within $1 million.
• Data rejections (detailed in Section 8): est netIncomeAvg and ebitdaAvg rejected as non-GAAP and non-comparable (FY2026E net income $21.328B against actual GAAP $17.087B, +24.8%; FY2026E EBITDA $26.792B against actual $32.138B, −16.6%) — epsAvg verified against the earnings calendar and used exclusively; capitalLeaseObligationsCurrent of zero flagged as implausible and the operating/finance split unverifiable because the 10-K's lease-maturity table was not captured in the extracted text; netDebt corrected from $124.900B to $124.295B and enterprise value from $544.642B to $544.037B (a 0.11% overstatement, the smallest in this batch); all book-value fields rejected (bookValuePerShareTTM $14.96 includes $4.954B of preferred and $548M of minority interest — common book value is $13.04; tangible common equity is approximately −$27.936B; priceToFairValueTTM of 9.866 is numerically identical to price-to-book and is not a fair-value estimate); returnOnEquityTTM of 50.6% rejected as a leverage artefact; all free-cash-flow multiples rejected (freeCashFlowYieldTTM −5.64%, priceToFreeCashFlowRatioTTM −17.72x, evToFreeCashFlowTTM −22.99x); trailing GAAP EPS adjusted by us from $5.83 to $4.61 for a $2.7B one-off gain and a 12.6% versus 19.9% tax rate, taking the trailing multiple from 24.99x to 31.6x; internal 52-week-range and moving-average conflicts between the quote and tech blocks documented and resolved in favour of quote for the range and tech for momentum; peers rejected as unusable (constituents from $19.5M to $659.8B against a $419.7B company); profile.ceo flagged as incomplete against a 10-K that refers to two Chief Executive Officers.
• Basis note — the most important one in this dive: the vendor payload contains no representation of $260 billion of off-balance-sheet lease commitments, $19 billion of post-year-end purchase commitments or a $3.3 billion lessor guarantee, all disclosed in the FY2026 10-K. These are not vendor errors — the items are correctly absent from a balance sheet — but every leverage ratio computable from the payload understates Oracle's forward obligation by roughly a factor of three. They are stated as undiscounted future lease payments over fifteen-to-nineteen-year terms and are not added to enterprise value.
• Cash-flow note: FY2026 free cash flow was negative $23.686 billion (operating cash flow $31.977 billion, capital expenditure $55.663 billion, both corroborated in the 10-K). The $5.787 billion dividend was funded from borrowings, as were the last $23.7 billion of the buildout. The current dividend yield of 1.37% should be read as a use of capital-markets proceeds, not as a return of surplus cash.
• Estimate coverage: deep on the near years and thin on the far ones — 32 analysts on FY2027 revenue and 26 on EPS; 30 and 24 on FY2028; 22 and only 11 on FY2029; 11 and 8 on FY2030. The FY2027 row is the primary forward anchor and the FY2028 row the secondary; the FY2029 to FY2031 rows are reported for shape and no conclusion rests on them. All estimates are non-GAAP and exclude stock-based compensation of $4.811 billion, or $1.65 a share — an SBC-inclusive alternative is shown alongside every forward multiple.
• Peer note: no peer-multiple comparison is drawn. The vendor peer set spans a $19.5 million company to a $659.8 billion company, and none of the plausible names (SAP, ASML, Adobe, Palantir, Synopsys, Fortinet) carries Oracle's leverage or capital intensity — which is the substantive point rather than a data complaint.
• Fair-value caveat: the $85 / $158 / $265 anchors are multiples of consensus non-GAAP EPS (10.5x / 19.6x FY2027E and 24.3x FY2028E respectively), cross-checked on enterprise value to revenue and restated on an SBC-inclusive basis. Stated scenario arithmetic, not a discounted cash flow. The base case assumes mild multiple expansion from 18.1x to 19.6x, which is flagged in Section 6b as the fragile leg. The bear case is deliberately not a distress multiple — a genuine credit event takes the equity well below $85, and that tail is acknowledged rather than modelled.
• Timing: the FY2026 10-K was filed 2026-06-22, forty-three days before this dive, and Q4 FY2026 results were released 2026-06-10. The next print is 2026-09-08, thirty-five days away — outside the five-day banner rule but squarely inside the decision window, and it is the pre-registered flip event. No earnings-call transcript is available on this plan, so there is no management commentary on backlog conversion, counterparty quality or financing plans beyond what the filings state; on this name that is the most consequential gap in the dataset.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.