Oracle

Refresh delta (since 2026-06-23)

The window carried no new filing and no new mega-contract — but two genuinely material things happened, both on the bear side of the ledger, and both sharpening the exact risks the prior dossier flagged as its top open questions.

What changed (new facts, all dated):

1. S&P NOTCH DOWNGRADE — BBB → BBB-, 2026-07-09. S&P Global cut Oracle's long-term issuer rating one notch to BBB-/A-3 (from BBB/A-2) — one step above speculative grade — retaining a stable outlook, and explicitly naming Oracle's concentration on OpenAI as a "central credit risk". The prior dossier (2026-06-23) said the rating risk was "rising even if the rating hasn't moved." It has now moved. This is the first actual notch cut (Sept-2025 was an outlook cut, not a downgrade).
2. OpenAI ≈ 50% of the $638B RPO — now a sourced number. In the downgrade, S&P estimates roughly half of the $638B remaining performance obligations is attributable to OpenAI (~$319B). This directly answers the prior dossier's #1 open management question ("what fraction of RPORemaining performance obligationsRevenue a company has already signed contracts for but has not yet delivered or booked. is OpenAI?") — previously unquantified. A single, un-IPO'd, not-yet-cash-generative counterparty is now confirmed as ~half the backlog.
3. S&P's funding-wall math is materially worse than the company's guide. S&P now models FY27 capex $90–95B (up from S&P's prior $60B assumption; above Oracle's own ~$70B net-capex guide ) and a FY27 free-operating-cash-flow deficit of ≈ −$42B (vs FY26 actual ≈ −$23.7B). The cash deficit is now projected to widen ~78%, not narrow.
4. The stock fell ANOTHER ~31% and set a fresh 52-week low. $166.27 (2026-06-23) → $114.99 (2026-07-24 close, −4.21% on the day; after-hours $114.85). That is ≈ −67% from the $345.72 52-week high and a new 52-wk low of $114.75. Market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth. $331.2B (from ~$503B a month ago). The mid-window path: 52-wk low ~$132 on ~2026-07-13 ("down 28% in a month" ), then lower into the S&P cut and the July 24 −4.27% session.
5. The multiple fully de-rated to an infrastructure/"show-me-the-cash" level. Forward P/E ≈ 14.3x, trailing ≈ 19.7x — the forward multiple is now below the software-industry median (~18x) and below the ~22x the prior dossier recorded a month ago. Analyst dispersion widened at both ends: CLSA cut to $145 on 2026-07-20 (new Street low) vs Guggenheim $400 (2026-06-11 high); consensus ~$265 across ~35 analysts.

What held (carried, unchanged):

• RPO = $638B and its conversion curve are unchanged (no new print until Sept). Only ~12% converts within 12 months (see Lens 5). The number is stale-but-current; the interpretation moved because the OpenAI share is now quantified.
• The structural thesis — a 49-yr database annuity with a debt-funded, OpenAI-concentrated GPU-landlord bolted on — is intact and, again, more sharply etched, not broken.
• No MarketCall is tracked for ORCL. Promotion to a position remains a separate, human-gated our position log step. This pass logs data and mechanics only; own view NOT ASSESSED.

Net one-liner (mechanics, not a call): In 32 days with no new filing, the two things that resolved both cut bearish — the credit rating actually dropped a notch, and the single biggest unknown (OpenAI's share of the backlog) resolved to "about half." The stock priced it: another −31%, a full de-rate to a sub-software multiple, at a fresh 52-week low.

---

Performance & RPO — re-run (the headline datapoint)

No new audited numbers this window (FY26 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. is the anchor; Q1 FY27 prints ~Sept 2026). The moving element is

RPO — size, growth, duration, concentration, cancellability

• Size: RPO $638 billion as of 2026-05-31 (vs $138B a year earlier — +362%). The 10-K attributes the jump "primarily [to] certain significant cloud contracts entered into during the period." QoQ it was +$85B off the fiscal-Q3 $553B.
• Duration (the decision-relevant curve, verbatim from the 10-K): Oracle expects 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". So ~88% of the $638B is dated beyond 12 months and ~20%+ ("the remainder") sits beyond 60 months. It is a long-dated promise, not near-term revenue.
• Concentration (NEW, sourced): ~50% of the $638B ≈ $319B is OpenAI, per S&P's downgrade rationale. Oracle itself does not disclose the customer split in the 10-K; this is the first credible external quantification, and it confirms the concentration the prior dossier could only assert.
• Cancellability — what the 10-K does and does NOT say: The RPO note contains no explicit cancellation-terms disclosure. Oracle discloses only that RPO "represent contracted revenues that had not yet been recognized" and that it "elected the optional exemption to not disclose the variable consideration" for contracts where variable consideration is allocated entirely to wholly-unsatisfied obligations — i.e., some usage/variable consideration is excluded from the $638B, and the fixed/minimum-commitment vs cancellable split is not broken out. The only cancellability signals are general risk-factor language: customers could "delay, decrease or cancel purchases" in a downturn, and a downgrade could "affect the terms or availability of certain long-term commitments (including data center leases), limit eligibility to contract with certain customers, and increase collateral, letter of credit or other credit support requirements". That last clause is newly live given the 2026-07-09 S&P cut. Flag for deeper review: the fixed-vs-cancellable composition of the OpenAI ~$319B is the single most important undisclosed fact in the file.

OCI / cloud growth, capex, FCF (carried from FY26 10-K; forward via S&P)

• OCI (cloud infrastructure) revenue $18,101M (+77% / +75% CC) FY26, up from $10,234M — the entire growth engine. Total cloud & software $58,530M (+19%); cloud = 51% of total revenue FY26. (Unchanged — carried.)
• Capex FY26 $55.7B (from $21.2B, +163%); the 10-K explicitly guides the "upward trend to continue during fiscal 2027 and in the following fiscal years". FCF FY26 ≈ −$23.7B ("FCF as % of net income: −139%").
• Forward capex/FCF (the moved number): company guide ~$70B net capex FY27; S&P now models $90–95B capex and −$42B FOCF for FY27. The deficit widens before it inverts — the "FCF waterfall" bulls model still sits in FY29–30, i.e. mostly in the months-37–60 bucket of the RPO curve.

---

Forensic Red-Flags & Credit — re-run (the window's real action)

Forensic lens; every figure labelled.

• Credit rating — the notch actually dropped. S&P BBB- / A-3, stable (cut from BBB/A-2 on 2026-07-09), one step above junk, OpenAI named "central credit risk". Moody's Baa2, negative outlook (further cut possible medium-term) — Baa2 sits one notch above S&P's new BBB-, so a Moody's cut would align them at the BBB- level. Fitch BBB / F2, stable — affirmed, and it rated Oracle's benchmark unsecured bonds BBB, but that Fitch action dates to 2026-02-02 (the $25B February deal), not the window. Net: two of three agencies now at the BBB-/negative-or-cut tier; the downgrade the prior dossier warned was "trending" has arrived at S&P.
• The cash-vs-earnings gap, now compounded by the funding forecast. GAAP net income FY26 $17,087M against FCF ≈ −$23.7B; S&P projects the FY27 FOCF deficit widening to ≈ −$42B on $90–95B capex. Debt was $129.5B aggregate at 2026-05-31, maturing 2026–2066; commentary now frames the load at **~$160B** ("$160B debt shadow"), above the last audited figure — treat $129.5B as the anchor and ~$160B as unverified pending the next filing.
• Off-balance-sheet commitments sharpen the "who pays the rent" picture (primary source, worth surfacing): beyond the $129.5B debt and $26.6B on-BS operating-lease liabilities, the FY26 10-K discloses $260 billion of ADDITIONAL lease commitments — "substantially all related to data center arrangements," expected to commence between Q1 FY27 and FY29, terms of 15–19 years, not reflected on the 2026-05-31 balance sheet, and including a lease under which Oracle guaranteed up to $3.3B of the lessor's borrowing (matures Sept 2026). Plus the carried $19B unconditional purchase commitment for cloud-infra assets entered subsequent to 2026-05-31 (commences FY27, 5-yr term) and unconditional purchase obligations primarily for data-center power. So against a $638B RPO of customer promises sits a stack of Oracle's own multi-year rent/purchase promises: $129.5B debt + $26.6B on-BS leases + $260B off-BS lease commitments + $19B purchase commitment + power obligations. This is the balance sheet of a leveraged infrastructure developer.
• Restructuring / operating stress (carried): FY26 $1.84B restructuring charge, ~21,000 jobs / ~13% cut (Revenue & Health Sciences ~−30%; OCI/AI shielded). Management signalled more to come; no new headcount datapoint surfaced this window.
• Litigation — scheduled events NOT confirmed this pass (flag): the FY26 10-K's securities class-action scheduling order had the amended complaint due 2026-07-14 and the Netherlands privacy Supreme Court judgment scheduled 2026-06-28. News search this pass surfaced no outcome for either — neither confirmed nor denied. Both remain open items to verify against the next 10-QForm 10-QThe quarterly version of the annual report. Lighter, and not audited. (Q1 FY27, ~Sept) or the dockets. The securities case is Barrows v. Oracle, D. Del., filed 2026-02-03, class period 2025-06-12 → 2025-12-16, naming Ellison + both co-CEOs.

---

Stock-Price Catalysts + Lens 7 · Valuation snapshot — re-run

Valuation snapshot (2026-07-24 close):

What the tape reacted to in the window (mechanics): not any beat (none due) — it traded on (1) the S&P downgrade, (2) the OpenAI-concentration quantification (~50% of RPO), (3) the widening FY27 cash-deficit forecast, and (4) generic AI-capex/valuation compression. The month's move (−31%) came with zero new operating data — pure re-rating. Forward P/E falling to ~14x means the market has now discarded the software multiple entirely and prices ORCL closer to a levered-infrastructure name; the prior dossier's "good news beats, stock falls" divergence has resolved into "no news, stock still falls," i.e. the de-rate is now valuation/credit-driven, not headline-driven.

Peer comps: still n/a for MSFT/AMZN/GOOGL/CRWV this pass (carried open across three refreshes; not fabricated).

---

Lenses carried forward unchanged (see the previous dossier)

• Lens 1 Business / Lens 4 Segments — three-segment reorg, FY26 audited mix (cloud 51% of revenue, OCI +77%, software support $19.8B flat annuity). Unchanged; no new print.
• Lens 2 Supply Chain — NVIDIA chokepoint + power constraint + customer-supplied-GPU structure; 181 data-center locations, substantially all leased. Unchanged (the $260B off-BS lease commitment above reinforces lease-intensity).
• Lens 3 Moats — genuine database annuity moat; OCI "moat" is a balance-sheet/willingness-to-lever edge, not tech/cost; segment margin already slipped 63%→59%. Unchanged.
• Lens 6 Earnings-call sentiment — "demand infinite → we can fund it → we'll need ~$20B more → cutting 13% to protect margin." No new call in window (next = Q1 FY27, ~Sept). Watch for a concrete FCF-turn year.
• Lens 9 Management — founder-controlled (Ellison ~40%+), co-CEO troika, Catz to Exec Vice Chair; all-in AI-capex legacy bet, buybacks dead ($95M), dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. levers live. Unchanged; the S&P cut is a new external check on the strategy.
• Lens 11 Projection / Lens 12 Bull-vs-Bear / Lens 13 Devil's Advocate / Lens 14 Management Questions — the scenario dispersion, the pre-mortem, and the 15 questions all stand; the OpenAI-≈50%-of-RPO datapoint makes questions #1, #2, #7 partially answered (concentration confirmed high) and #3, #4, #8 (FCF path, sources-and-uses, downgrade thresholds) more urgent post-cut. No re-run needed — the prior text is intact.

---