Web Research

Web Research — What the Internet Knows That the Filings Don't

Bottom line. Between the FY2025 10-K and today (2026-06-20), Oracle reinvented itself as the debt-financed landlord of the AI compute boom — and the public record reveals the one thing the filings cannot: the market has already turned. ORCL trades near $184, down roughly 46% from its September 2025 all-time high of $345.72, despite posting record results every quarter, because investors have stopped paying for the $638B backlog and started demanding proof it converts to cash. The web supplies the two facts that move the thesis hardest and sit outside any filing: (1) roughly half of that backlog is tied to a single, cash-burning counterparty — OpenAI's ~$300B/5-year deal — with no disclosed credit allowance, and S&P warns OpenAI could be over a third of Oracle's revenue by FY2028; and (2) a securities-fraud class action (filed Feb 3, 2026) now alleges management oversold the AI-capex payoff while free cash flow went to negative $23.7B. The growth is real and filing-confirmed; the debate is entirely about cash conversion, counterparty risk, and a balance sheet now carrying well over $100B of debt.

This briefing leads with what changes the stock, not with what each source said. Web facts carry their outlet and URL; raw filing facts I introduce to confirm or contradict them carry a numbered page citation.

---

The single most important picture: a record backlog, a collapsing stock

Source: RPO ended Q4 FY2026 at $638B, up 363% YoY and $85B sequentially [1]; prior quarters per Oracle releases and CNBC earnings coverage (Sep 2025 RPO $455B, +359%: https://www.cnbc.com/2025/09/09/oracle-orcl-q1-earnings-report-2026.html).

Sources: price ~$184 / 52-week range $134.57–$345.72 (Yahoo Finance, Jun 2026, https://finance.yahoo.com/quote/ORCL/; Morningstar, Jun 18 2026, https://www.morningstar.com/stocks/xnys/orcl/quote); consensus 12-month target ~$262 (TipRanks, https://www.tipranks.com/stocks/orcl/forecast).

The tension above is the whole story: the backlog (left) keeps setting records while the equity (the ~46% drawdown) keeps de-rating. Everything below explains why, ranked by how much it should move an investor's view.

---

The findings, ranked by importance

1. The AI premium has been violently re-rated — record results no longer move the stock up

ORCL has fallen ~46% from its September 2025 peak (~$345) to ~$184, even though Q4 FY2026 (reported June 10, 2026) was a record: revenue $19.2B (+21%), cloud +47%, IaaS +93%, non-GAAP EPS $2.11 [1]. The stock still fell more than 10% on the print (INDmoney, Jun 11 2026, https://www.indmoney.com/blog/us-stocks/oracle-q4-fy2026-earnings-orcl-stock-drop), and TheStreet noted "the problems which precipitated its nearly 40% decline from its Sept. 2025 all-time high are alive and well" (https://www.thestreet.com/latest-news/orcl-oracle-earnings-call-updates-q4-2026).

So-what: The burden of proof has flipped from narrative to delivery. The multiple compression — not any operational miss — is the dominant driver of the loss, so the risk/reward now hinges on FCF inflection and backlog conversion, not on bookings optics. Priced in? The direction (AI-infra de-rating) is largely in the price; what is unresolved is whether margins and cash inflect before the financing strain bites — that is the swing factor. Red flag on the chart; the de-rate itself is the opportunity if the bull case holds.

2. ~Half the $638B backlog is OpenAI — the single biggest concentration risk, with no disclosed allowance

Multiple independent sources size OpenAI's commitment at roughly $300B over five years from 2027 (~$60B/yr, 4.5GW) — close to half of total RPO (SiliconANGLE citing WSJ, Sep 10 2025, https://siliconangle.com/2025/09/10/openai-oracle-strike-300b-cloud-computing-deal-power-ai). One analyst estimate puts OpenAI at ~54% of RPO; the top three customers are reportedly over 60% of AI contracts (Investing.com, https://www.investing.com/analysis/oracle-backlog-of-553b-raises-questions-around-future-revenue-scale-200679538). S&P Global warned on Sep 24, 2025 that OpenAI could exceed one-third of Oracle's total revenue by FY2028 (via Levi & Korsinsky, https://zlk.com/cases/oracle-corporation-class-action-lawsuit-orcl). OpenAI's own ~$25B ARR (early 2026) is dwarfed by its ~$60B/yr compute bill, and Oracle has disclosed no allowance or credit reserve against this RPO, arguing it can reshift GPU capacity "within hours" if a customer fails to pay (Motley Fool, Jan 7 2026, https://www.fool.com/investing/2026/01/07/oracles-secret-weapon-against-ai-customer-risk-has).

So-what: OpenAI's creditworthiness has become Oracle's revenue-recognition and collectibility risk. If a marquee AI customer slips, both the revenue ramp and the debt-service math break at once — this is the asymmetric downside in the thesis and the reason the backlog no longer commands a premium. Priced in? The concern is now mainstream and is much of why the stock halved, but it is unresolved — Oracle does not disclose customer-level RPO, so the market is trading on analyst estimates, not disclosure. Red flag.

3. Free cash flow is deeply negative and the buildout is debt-funded — the crux of the bear case

Oracle's own Q4 release confirms FY2026 free cash flow of negative $23.7B, even as operating cash flow rose 54% to a record $32.0B [1]. The gap is capex: FY2026 capex ran ~$55.7B and management guides FY2027 to roughly $90–95B gross (~$70B "net" of customer prepayments) (Global Data Center Hub, Jun 19 2026, https://www.globaldatacenterhub.com/p/oracle-q4-fy2026-hyperscaler-capex). To fund it, Oracle raised $43B of debt and $5B of equity in FY2026 and plans ~$40B more in FY2027, including a $20B at-the-market equity issuance [2]. Total debt is now reported at roughly $108–135B (The Motley Fool, Dec 11 2025, https://www.fool.com/investing/2025/12/11/oracles-debt-balloons-to-108-billion-as-ai-spendin/).

Sources: FY26 operating cash flow $32.0B and free cash flow negative $23.7B [1]; FY26 capex ~$55.7B and FY27 ~$90–95B gross (Global Data Center Hub, Jun 19 2026, https://www.globaldatacenterhub.com/p/oracle-q4-fy2026-hyperscaler-capex).

So-what: Oracle no longer self-funds its growth — every incremental GW of capacity comes from debt and dilution. Equity issuance dilutes; the debt raises interest expense against a 'BBB'/'Baa2' rating only two notches above junk. The equity story now depends on FCF inflecting positive (Mizuho models ~2029) as contracted capacity comes online. Priced in? This is the principal cause of the selloff, so largely recognized — but a sustained negative-FCF print or a downgrade would pressure the multiple further. Red flag.

4. A securities-fraud class action now overhangs the stock, with alleged heavy CFO selling

Barrows v. Oracle (Case No. 1:26-cv-00127, D. Del.) was filed February 3, 2026, with a class period of June 12 – December 16, 2025 and an April 6, 2026 lead-plaintiff deadline; defendants include Ellison, Catz, co-CEOs Magouyrk and Sicilia, and CFO Kehring (Kessler Topaz/PRNewswire, https://www.prnewswire.com/news-releases/oracle-corporation-orcl-investors-april-6-2026-deadline-in-securities-fraud-class-action-lawsuit-filed-by-kessler-topaz-meltzer–check-llp-302720249.html). The complaint alleges defendants touted that AI capex would "quickly" drive revenue while concealing risks to debt, the credit rating, and FCF — and that Safra Catz sold nearly 8.7 million shares for over $1.82 billion during the class period, more than double the prior comparable period (Levi & Korsinsky, https://zlk.com/cases/oracle-corporation-class-action-lawsuit-orcl). A second wave of investigations (SueWallSt, Schall, Glancy Prongay) launched after the June 2026 Q4 print (PRNewswire, Jun 11 2026, https://www.prnewswire.com/news-releases/oracle-corp-investigation-initiated-suewallst-investigates-the-officers-and-directors-of-oracle-corp-orcl-302798496.html).

So-what: Active litigation conveniently catalogs the entire bear case and the stock-drop dates, raising legal-cost and headline risk; the alleged accelerated CFO selling concurrent with bullish guidance is the kind of fact that survives a motion to dismiss and weighs on management credibility. Priced in? The litigation is public but its merits are unresolved — a tail risk the market has noted but not quantified. Red flag.

5. The 30–40% OCI margin promise is unproven — internal docs suggest mid-teens during the ramp

Management frames a 30–40% OCI margin as a target reached only once capacity is online, not a current reality (Computer Weekly, Dec 11 2025, https://www.computerweekly.com/news/366636165/Oracle-expects-to-increase-OCI-margins-by-30-40). The Information's leaked internal documents reportedly showed Oracle earning only ~14–16% renting Nvidia chips, with a ~$100M loss on early Blackwell rentals (DataCenterDynamics, https://www.datacenterdynamics.com/en/news/internal-documents-show-oracles-average-profit-margin-for-ai-cloud-deals-was-16-report/). Oracle counters that Q3 FY2026 delivered-capacity gross margin was ~32%. Consolidated gross margin has stepped down from 72.5% (May 2024 quarter) to 64.6% (Feb 2026 quarter) as low-margin infrastructure scales (TIKR, https://www.tikr.com/blog/oracle-stock-q4-2026-beats-estimates-as-oci-revenue-surges-93-year-over-year).

So-what: Return-on-capex is the valuation crux. If AI-cloud margins settle in the mid-teens rather than 30%+, the math on ~$90B/yr of capex and the FY2030 revenue map ($144B+ targets) does not work, and the de-rate is justified rather than overdone. Priced in? Bears focus on the leaked mid-teens figure; bulls on the 32% claim — the truth is genuinely contested and will only be settled by reported segment economics as capacity ramps. Red flag / watch.

6. Accounting quality flags: a server-life extension and a one-time gain are flattering reported earnings

Oracle extended the useful life of servers and networking equipment from five to six years effective the start of FY2025, which decreased operating expense by $733M and increased net income by $573M (+$0.21 EPS) in FY2025 alone — on top of a 2023 extension from four to five years [3]. This runs directly against Amazon, which cut server life to five years effective Jan 2025 citing faster AI obsolescence (Bloomberg via Yahoo, Feb 12 2025, https://finance.yahoo.com/news/meta-accounting-move-ai-servers-124059775.html). Separately, Q2 FY2026 GAAP and non-GAAP EPS were both boosted by a $2.7B pre-tax gain on the sale of Oracle's Ampere stake [4].

So-what: Oracle is lengthening depreciation precisely as the industry shortens it — boosting near-term profit while GPUs arguably age faster, not slower. Combined with a one-time gain and a persistent ~30% GAAP-to-non-GAAP EPS gap (driven by SBC and acquisition amortization), the "record" headline numbers overstate economic profit just as the cash story deteriorates. Priced in? Largely overlooked by the consensus, which anchors on non-GAAP EPS — an under-appreciated quality-of-earnings drag. Red flag.

7. Credit markets blinked, then recovered — but a key project financier walked away

Oracle's CDS spiked to its widest since 2009 in late 2025 on debt-load fears, then tightened sharply after the Feb 1, 2026 financing plan reassured bond investors; Fitch affirmed 'BBB'/Stable on Feb 2, 2026 and Moody's affirmed Baa2 (CNBC, Feb 2 2026, https://www.cnbc.com/2026/02/02/oracles-credit-default-swaps-are-plummeting-after-financing-plans.html). But the strain is visible: on Dec 17, 2025 the Financial Times reported Blue Owl Capital backed out of a $10B financing for an Oracle data center serving OpenAI, over concerns about Oracle's spending and debt (via Kessler Topaz, https://www.prnewswire.com/news-releases/oracle-corporation-orcl-investors-april-6-2026-deadline-in-securities-fraud-class-action-lawsuit-filed-by-kessler-topaz-meltzer–check-llp-302720249.html), and a December 10-Q disclosed ~$248B of additional lease commitments that CreditSights called a "bombshell."

So-what: The IG affirmation removes the tail risk of a downgrade-driven funding spiral — a genuine positive that bonds rallied on. But a marquee financier walking and $248B of off-balance-sheet leases show the external capital underwriting the buildout is neither free nor guaranteed; the 'BBB' cushion is thin if FCF stays negative. Priced in? Credit relief is recognized (CDS rally); the off-balance-sheet lease scale is less appreciated. Neutral, leaning red.

8. The genuine positive: the database moat is travelling into rivals' clouds — but OCI's share is still tiny

The cleanest evidence the franchise is extending rather than eroding is multicloud database (Oracle DB running inside Azure/AWS/Google): revenue grew 404–531% YoY through FY2026, and Ellison cited 1,529% growth at the analyst meeting, targeting ~$20B in five years (Oracle CEO blog, Mar 24 2026, https://blogs.oracle.com/ceo/from-the-q3-earnings-call). Against that, OCI's infrastructure share remains only ~3–5% (#4–5 globally) versus AWS ~30%, Azure ~20%, Google ~13% — flat for years despite +93% growth, because the base is small (CRN/Synergy, Nov 6 2025, https://www.crn.com/news/cloud/2025/global-cloud-market-share-q3-2025-aws-lowers-microsoft-and-google-stay-same). On core databases, Gartner now ranks Oracle #3 (behind AWS and Microsoft since 2022), with new builds defaulting to PostgreSQL (The Register, Apr 21 2026, https://www.theregister.com/2026/04/21/gartner_bamboozling_dbms_chart_shows/).

Source: Synergy Research via CRN, Nov 6 2025 (https://www.crn.com/news/cloud/2025/global-cloud-market-share-q3-2025-aws-lowers-microsoft-and-google-stay-same); OCI ~3–5% range, figure shown at 4%.

So-what: "Distant fourth, gaining" is accurate on growth but aspirational on share — the thesis is a bet on share capture, not incumbency. Multicloud is the highest-quality, most defensible part of the story (it monetizes the moat without forcing customers off rival clouds); commodity IaaS share is not. Priced in? Multicloud's scale is under-appreciated by bears; OCI's small absolute share is under-appreciated by bulls. Positive (multicloud) / neutral (share).

9. Governance: Ellison's pledged stake and ~42% control are the structural overhang

Larry Ellison pledged 346 million shares (~30% of his stake) as of Sep 19, 2025 "to secure certain personal indebtedness" — up from 277M (24%) a year earlier — under a carve-out created specifically for him after a 2018 ban on other insiders pledging (IFR/Refinitiv, Oct 4 2025, https://www.ifre.com/topic-codes/2321703/larry-ellison-pledges-30-of-his-oracle-shares-in-loans). Ellison owns ~42% of Oracle and remains Executive Chairman and CTO, giving him effective control (Wikipedia, citing DEF 14A/10-K, https://en.wikipedia.org/wiki/Oracle_Corporation). Insiders are net sellers with no notable buying; the new co-CEOs both recorded sales near their promotions (TIKR, https://www.tikr.com/blog/who-owns-oracle-biggest-shareholders-and-recent-insider-trades).

So-what: A ~42% founder stake plus a pledged block creates forced-sale risk in a sharp drawdown (a margin call could pressure the stock precisely when it is already falling) and concentrates power such that minority holders cannot force change — the classic controlled-company governance discount. Priced in? Oracle calls the pledge "not a material risk"; the market has not priced a forced-sale scenario. Red flag (mild, tail risk).

10. Leadership transition mid-supercycle: new co-CEOs, new CFO, and ~12,000 layoffs

On Sep 22, 2025, insiders Clay Magouyrk (ex-OCI president) and Mike Sicilia became co-CEOs, with Safra Catz moving to Executive Vice Chair; the stock rose ~6% on continuity signaling, and the two received $250M/$100M retention grants (Reuters, https://www.reuters.com/sustainability/boards-policy-regulation/oracle-appoints-insiders-clay-magouyrk-mike-sicilia-co-ceos-2025-09-22/). A new external CFO, Hilary Maxson, was hired in April 2026, and Oracle laid off nearly 12,000 staff in India in April 2026 (Wikipedia citing The Hindu BusinessLine, https://en.wikipedia.org/wiki/Oracle_Corporation).

So-what: A two-CEO structure under a still-dominant founder-Chairman concentrates real authority in Ellison and creates coordination/accountability risk during Oracle's largest-ever capital cycle; CFO turnover plus mass layoffs during the transition signal cost pressure, not just continuity. Priced in? Treated as continuity by the market — under-watched as an execution risk. Neutral, leaning red.

11. Consensus is still bullish and contrarian to the price — where the disagreement lives

Despite the crash, the sell side remains constructive: ~22 Buy / 5 Hold / 0 Sell, 12-month average target ~$262 (TipRanks, https://www.tipranks.com/stocks/orcl/forecast), but with an extraordinary spread of $155 to $400 (CNN, https://www.cnn.com/markets/stocks/ORCL). Estimate revisions are net positive, especially revenue (Seeking Alpha: 30 up / 3 down over three months, https://seekingalpha.com/symbol/ORCL/earnings/revisions).

So-what: The ~$262 consensus versus ~$184 spot implies ~40% upside and zero Sell ratings — either the targets lag the de-rate, or the market is mispricing a still-intact growth story. The $155–$400 spread is the thesis: it quantifies how unresolved the AI-capex bet remains. Priced in? The bullish consensus is contrarian to the tape — the gap is the edge, in either direction. Neutral / positive.

---

Recent-news reference layer

Meaningful news from roughly the last three months, extended back to capture still-live events (the OpenAI deal, Stargate, the EBS breach, the Cerner/Epic erosion, the securities litigation). This is the reference table the findings above are drawn from.

Sources: corpus news file (curated headlines, 2023–2026) [5] plus the individual outlet URLs cited inline in the findings above; RPO/financials confirmed in the Q4 FY2026 8-K [1] and the $2.7B Ampere gain in the Q2 FY2026 8-K [4].

---

Other signals worth flagging

---

Specialist question coverage

The thesis-changing specialist answers are promoted into the ranked findings above (RPO/OpenAI concentration → #2; OCI margins → #5; capex/FCF/credit → #3 and #7; Ellison pledge and co-CEO → #9 and #10; multicloud and DB share → #8). The remainder, with synthesized answers and confidence, is below.

---

Where the real uncertainty still sits

Three threads no source in the corpus, the web, or my searches fully resolved, and where the PM's edge remains: (1) the exact OpenAI share of RPO and any credit support — Oracle does not disclose it, so the ~50% figure is an estimate the market is trading on; (2) the true steady-state AI-cloud margin — the ~14–16% leaked figure versus the 32% claim is the single biggest swing factor and will only be settled by reported segment economics; and (3) the FY2025 say-on-pay vote and proxy-advisor stance on the co-CEO mega-grants and Ellison pledge — absent from the dataset and worth pulling from the 2025 proxy directly. On everything else, the public record is unusually rich and the bear case is well-documented rather than hidden — which itself tells a PM the risks are largely known, and the question is whether they are now over- or under-priced at a ~46% drawdown.