Oracle

Phase A — Understand the business

Lens 1 · Company Overview — re-run (segment reorg + audited mix)

Oracle is a 49-year-old enterprise-software company (founded 1977, HQ Austin TX, ~2.88B shares) that in ~24 months bolted a hyperscale AI-infrastructure landlord onto a mature database-and-applications franchise. The FY26 10-K formalizes a renamed three-segment structure — cloud and software (formerly "cloud and license"), hardware, services — each a single operating segment, with the CODMs being "our Chief Executive Officers and Chief Technology Officer". The three-CODM framing is itself a tell: the co-CEO/CTO troika now formally runs resource allocation.

FY26 revenue by business:

• Cloud and software $58,530M (+19%/+17% CC) — 87% of total.
• Hardware $3,084M (+5%) — 5%, finally growing slightly on Exadata/strategic hardware.
• Services $5,743M (+10%) — 8%, up on consulting.

Within cloud and software, the audited offering split is the whole story:

• Cloud applications $15,888M (+11%/+10% CC)
• Cloud infrastructure (OCI) $18,101M (+77%/+75% CC) — up from $10,234M; now the #2 line and closing on software support.
• Software license $4,737M (−9%) — the structurally declining perpetual-license book.
• Software support $19,804M (+1%) — the legacy annuity, ~flat, still the single largest revenue line.

Cloud now = 51% of total revenue (FY26) vs 43% (FY25) — the audited annual number; the prior dossier's "52% cloud" was the Q3 exit rate. The structural shift is real and accelerating, and it is entirely an OCI story (cloud infra is 84% of cloud growth in CC).

Contract structure — the crux, unchanged. The new book is large, multi-year, take-or-pay capacity contracts with a handful of AI labs and neoclouds, with the unusual customer-prepay / customer-supplied-GPU feature. The 10-K now quantifies the prepay: operating cash flow included $4.6B of customer prepayments with a significant financing component in FY26 — i.e., customers are pre-funding capacity, which both de-risks Oracle's capex and deepens counterparty entanglement.

Customers: the legacy installed base for the annuity; for the new book, OpenAI (via Stargate — $60B/yr for five years, 2027–2031 = $300B ), plus reported xAI, Meta, and US federal. Suppliers: NVIDIA (gating input), AMD, the data-center construction/power chain. Competitors: AWS, Azure, GCP; CoreWeave/Nebius/Crusoe; SAP/Salesforce/Workday in apps.

Lens 2 · Supply Chain — carried forward (unchanged)

Unchanged from deep-dive-2026-06-18.md. Map: NVIDIA GPUs (single most critical input; AMD secondary) → HBM + optical/networking (SK Hynix/Micron/Samsung; NVIDIA/Broadcom/Arista) → data-center shell + power (the binding 2026 constraint) → Oracle operates OCI regions (now 181 data-center locations, substantially all leased ) → end customer (OpenAI/Stargate, xAI, Meta, enterprises, US federal). Chokepoints: NVIDIA allocation (no Oracle proprietary silicon vs Google TPU / Amazon Trainium); power/interconnect (Stargate sized in GW, ~7GW planned ); and the customer-supplied-GPU structure that compresses Oracle's role toward operator. One new data point reinforcing the lease intensity: operating lease liabilities (non-current) more than doubled to $26,648M from $11,536M — the data-center lease book is ballooning alongside owned capex.

Lens 3 · Competitive Advantages (moats) — carried forward, one new stress point

Thesis unchanged: the legacy database moat is genuine (extreme switching costs, ~annuity support renewals — software support $19.8B, ~flat, proves the stickiness even as license sales decline). The OCI "moat" is thinner than bulls claim — it is a balance-sheet edge (willingness to lever up and commit capacity ahead of demand), not a technology or cost edge. New stress point from the audited numbers: the cloud-and-software segment margin compressed to 59% (from 63%) as cloud/infrastructure expenses rose +56% against +19% revenue — i.e., the AI buildout is actively diluting the blended cloud margin, exactly the commoditization-risk vector the bear case flags. "A moat made of debt-funded capacity is not a moat; it is a bet" stands. ``

Lens 4 · Segments — re-run (audited FY26)

Revenue by business, audited:

By geography: Americas $44,478M (+22%, now 66% of total), EMEA $15,297M (+9%/+3% CC), Asia Pacific $7,582M (+8%). The mix is tilting harder toward the Americas as the US AI buildout dominates growth (Americas = 88% of CC revenue growth). GAAP operating margin 31% ($20,606M op income), flat YoY — revenue growth funded the opex increase but did not expand margin, because infrastructure expense scaled with it.

The cause of acceleration is unambiguous and unchanged: OCI is the entire story. The new wrinkle is that the annuity (software support) is now dead flat — the legacy book is no longer growing at all; it is a stable cash base, not a contributor to the growth rate.

---

Phase B — Measure performance

Lens 5 · Earnings Result — re-run (FY26 now audited from the 10-K)

The prior dossier analyzed the fiscal Q4/FY26 print from the June-10 press release . The **FY2026 10-K (filed 2026-06-22)** now makes the full year :

• Revenue $67,357M (+17%/+16% CC). (Q4 standalone was +21% to $19.2B per the release.)
• GAAP net income $17,087M (vs $12,443M FY25, $10,467M FY24) — +37%. Note a one-time $933M net tax expense from the "One, Big, Beautiful Bill Act" (signed 2025-07-04) — i.e., underlying pretax was even stronger.
• Net income available to common $16,984M — the ~$103M gap is the new preferred dividend.
• Diluted GAAP EPS $5.83 (vs $4.34, $3.71) — +34%. Basic $5.94.
• What drove it: cloud +19% with OCI +77% full-year; legacy flat-to-down. Cloud = 51% of revenue.
• Operating cash flow $31,977M (+54%) — 187% of net income; helped by $4.6B of customer prepayments.
• Capex $55.7B (from $21.2B) — +163%. → FCF ≈ −$23.7B (10-K: "FCF as % of net income: −139%"). The prior `` −$23.7B is confirmed to the dollar.
• Balance sheet: notes payable & other borrowings $129.5B aggregate (non-current $122,342M + current $7,199M), up from ~$92.6B; operating-lease liabilities $26,648M. Funded in FY26 by $42.7B senior notes (net), $5.0B Mandatory Convertible Preferred, and $4.9B from selling the Ampere stake; dividends paid $5.8B; buybacks just $95M (confirming buybacks are dead).
• Guidance/tone: FY27 revenue guided +34% CC, non-GAAP EPS ~$8.05, net capex ~$70B.
• Market reaction — the tell, now twice over. Stock fell ~8.5% on the June-10 print (on the funding disclosure), then a further ~2% on June 23 on the layoff disclosure, to ~$166. The market is no longer paying for backlog; it is pricing funding, counterparty, and now operating-stress signals.

The NEW operating fact — a 13% workforce cut. FY26 headcount fell from ~162,000 to 141,000 — ~21,000 jobs, ~13% — the largest tech workforce reduction of the year, driving the $1.84B restructuring charge (vs $374M). The 10-K and reporting attribute it to AI-driven efficiency + restructuring; Revenue and Health Sciences cut ~30%, while OCI/AI teams were shielded or expanded; management signals more to come. Read two ways: (1) bullish-for-margin — Oracle is taking out a high opex base just as it needs every dollar for the FCF turn; (2) bearish-tell — a company gutting its non-AI workforce 30% while betting the balance sheet on AI is all-in with no hedge, and the Health Sciences (Cerner) cut signals that the $28B Cerner acquisition's growth case has faded.

Lens 6 · Earnings Calls (sentiment trend) — re-run (delta)

No transcripts on disk; ``. The trajectory from the prior dossier holds and extends: H2-2025 "demand is infinite" → H1-2026 "and we can fund it without dilution" → June-2026 "...we'll need ~$20B more, but the backlog is real." The FY26 10-K + layoff news add a fourth beat: "and we're cutting 13% of the company to protect margins." The recurring defensive line — "we don't expect to raise incremental funds for these contracts" — is now in direct tension with the $19B post-quarter purchase commitment and the ~$70B FY27 capex guide. Tone on demand stays confident; the market's repeated negative reactions to every disclosure say the reassurance is not landing. Watch the Q1 FY27 call (~Sept 2026) for whether management gives a concrete FCF-turn year — its continued absence is itself the signal.

Lens 7 · Comps — re-run (ORCL marks updated; peers still not pulled)

ORCL as of 2026-06-23:

Lens 8 · Stock-Price Catalysts (moves >5%) — re-run (new prints added)

Pattern (mostly ``): Sept 2025 RPO/Stargate euphoria → peak ~$302 (52-wk high $345.72); Feb 2026 AI-debt jitters + OpenAI-shortfall reports → trough $154, securities suit filed; Mar 2026 (Q3 print) beat + RPO $553B → −8.5% on capex fear; June 10, 2026 (Q4 print) beat + RPO $638B → −8.5% on the funding disclosure; June 23, 2026 −2% on the 21k-layoff disclosure → **$166**.

What the tape reacts to (sharpened): not the beat — the demand story is fully priced. It now trades on (1) RPO acceleration vs plateau, (2) any OpenAI/counterparty stress, (3) the funding plan, and now (4) operating-stress signals (layoffs). The stock has round-tripped from ~$302 to ~$166 — ~−52% from the peak — while every quarterly result beat. That divergence is, again, the single most important fact in this dossier, and it widened since the prior refresh: good news has not just stopped working; it is now actively met with selling.

---

Phase C — Judge people & books

Lens 9 · Management — re-run (layoffs = a new capital/operating-allocation datapoint)

Carried thesis: founder-controlled (Ellison ~40%+, Chairman/CTO), the AI-capex bet is his late-career legacy swing; Sept-2025 transition moved disciplined operator Safra Catz to Executive Vice Chair and installed co-CEOs Clay Magouyrk (OCI) and Mike Sicilia (Industries/apps) — institutionalizing the AI-infra bet right as FCF went negative. New datapoint: the 21,000-person/13% cut is now the clearest capital-allocation signal of the new regime — management is willing to gut the non-AI workforce (Revenue, Health Sciences ~−30%) to protect margins for the AI build, shielding OCI/AI. It is decisive and consistent with the all-in thesis; it is also a confession that the legacy organization was over-staffed for a flat-growth annuity and that the Cerner/Health Sciences growth case has stalled. Capital-allocation pivot is unchanged and extreme: capex ~$7B (FY24) → $21.2B (FY25) → $55.7B (FY26 actual) → ~$70B (FY27 guide), buybacks dead ($95M), debt $129.5B, plus the mandatory convertible (dilution). Red flags: the live OCI securities class action naming both co-CEOs + Ellison; the dilution levers; a single decision-maker with ~40% control and no effective check, swinging for legacy late in his career. Archetype unchanged: high-conviction, high-agency, low-check founder bet.

Lens 10 · Forensic Red Flags — re-run (audited FY26 + sharpened litigation calendar)

Forensic lens. Every figure labeled.

• Cash flow vs earnings — the big one, now audited. GAAP net income $17,087M while FCF ≈ −$23.7B. The gap is capex ($55.7B) — real, not an accrual trick — but it means non-GAAP EPS growth is celebrated while the company burns ~$24B and issues $42.7B of notes + $5B preferred to do it. SBC total $4,811M FY26 ($1,618M segments + $3,193M R&D/G&A) vs $4,674M FY25 — flattering non-GAAP; 10-K discloses annualized dilution ~1.0%/yr since June 2023 and max potential dilution 3.5%.
• RPO quality — the audited conversion curve is the bear case in one table. $638B RPO, of which the 10-K states Oracle expects to recognize ~12% over the next 12 months, ~34% in months 13–36, ~34% in months 37–60, and the remainder thereafter. So ~88% of the backlog is dated beyond a year and ~32% beyond five years — it is a long-dated promise from a few counterparties, not near-term revenue. A backlog is only as good as the payer over a 5-year horizon.
• Leverage / liquidity. Total debt $129.5B (non-current $122,342M) maturing 2026–2066; operating-lease liabilities $26,648M (>2x YoY); plus a NEW $19B unconditional purchase commitment entered after year-end. The funding stack is debt + preferred + leases + purchase commitments — increasingly the profile of a leveraged infrastructure developer, not a software company.
• One-time items to normalize: $1.84B restructuring (layoffs) and $933M tax charge (OBBBA) both depress GAAP FY26 — i.e., underlying earnings power is modestly higher than the headline, a point bulls will (fairly) make.
• Goodwill/intangibles: large legacy Cerner goodwill; the ~30% Health Sciences headcount cut raises (does not yet trigger) a future-impairment watch.

Regulatory findings (required sub-section) — regulatory/regulatory-findings.md (fetched 2026-06-18; <7 days old, not re-fetched per Step-0 rule) shows no SEC Litigation Releases and no AAERs naming Oracle, 2021–2026. The material legal items, now from the FY26 10-K Note 15 (Legal Proceedings) rather than the Q3 10-Q:

• Securities Class Action re: Oracle Cloud Infrastructure — filed 2026-02-03, U.S. District Court, D. Delaware, against Oracle, the CTO (Ellison), both co-CEOs, two other executives and one board member, alleging false/misleading statements about the cloud-infrastructure business. NEW scheduling order: amended complaint due 2026-07-14; defendants respond 2026-09-16; any motion-to-dismiss fully briefed 2026-12-18 (→ ruling early-mid 2027). Oracle "believes it has meritorious defenses." This is still the single most important legal item — the market is literally litigating the OCI growth narrative, and the calendar is now concrete.
• Netherlands Privacy Class Action (The Privacy Collective, Amsterdam, co-defendant Salesforce) — materially advanced: an Advocate General non-binding opinion (2026-01-30) advised the Dutch Supreme Court to dismiss Oracle's appeal and uphold TPC's in part; Supreme Court judgment scheduled 2026-06-28. Oracle still deems it immaterial, but it is now live within days.
• Non-SEC enforcement (web): no new material federal enforcement settlement surfaced this pass beyond known historical ad-tech/privacy matters ``.
• Credit: S&P BBB / negative, Moody's Baa2 / negative (cited OpenAI counterparty concentration), Fitch BBB / stable — unchanged from the prior dossier. The Sept-2025 "Moody's downgrade" headlines were an outlook cut to negative, not a notch downgrade (rating stayed Baa2). No new June-2026 notch surfaced. But the FY26 10-K confirms debt +$37B against flat operating margin, pushing leverage toward the ~4x downgrade trigger S&P/Moody's flagged — the rating risk is rising even if the rating hasn't moved.

Net: no SEC fraud findings; a live, now-scheduled securities class action squarely on the cloud narrative; an audited negative-FCF/$129.5B-debt profile; three agencies on negative-or-cautious outlook; and leverage trending toward the downgrade trigger. Clean-books verdict unchanged: not fraudulent, but the quality of earnings (cash vs GAAP, 88%-long-dated backlog, single-counterparty concentration) is the legitimate forensic concern.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (non-GAAP EPS, FY27–FY29) — re-run (anchored on audited FY26)

Anchored on audited FY26. FY26 actual non-GAAP EPS $7.63 / GAAP $5.83. Management FY27 guide: revenue +34% CC, non-GAAP EPS ~$8.05. Consensus FY27 EPS ~$8.1 (wide range).

Dispersion remains enormous because the swing factor is binary: does the $638B backlog convert at the contracted pace and price (the audited 12%/34%/34%/remainder curve says most of the proof is years out), or does the largest counterparty stress? The base case essentially is the company's guide; independent value-add is sizing the bear, not refining the base.

The cleaner forward question than EPS remains the FCF turn: FY26 FCF ≈ −$23.7B (audited); FY27 capex guided ~$70B against $67B→$90B revenue means the deficit likely widens before it inverts — bulls model a "free-cash-flow waterfall" in FY29–30 as capex rolls off and contracted revenue recognizes. The whole bull case is a duration bet: survive 2–3 more years of cash burn ($24B+/yr) and a $129.5B+ debt load to reach a waterfall that, per the audited conversion curve, mostly lands in months 37–60. Brier forecast not logged (unattended --watchlist rule — only log on genuine committed conviction; no forecast.ts create).

Lens 12 · Bull vs Bear — re-run (does it still hold given audited FY26?)

Bull case. Oracle has pre-sold $638B of cloud capacity — several times its trailing revenue — and is the only legacy enterprise vendor that reinvented itself into a top-tier AI-infra provider. The audited FY26 proves the inflection: revenue +17%, OCI +77%, op cash flow +54% to $32B, and the company simultaneously took out 21,000 jobs to protect margin. The database annuity ($19.8B support, sticky) funds the bet. If even ~80% of RPO converts near plan, FY29–30 FCF inflects violently positive and the stock re-rates on a cash-flow basis it has never carried. The customer-prepay structure ($4.6B of financing-component prepayments in FY26 alone) de-risks the capex more than bears admit. Ellison is doing late-career what he's done three times before. At ~$166 (−52% from peak) you buy the audited backlog at barely above the Feb trough and a forward P/E (~22x) that has nearly converged with the software median — while fundamentals only improved.

Bear case (2–3 permanent-impairment risks), all reinforced by the 10-K.

1. Counterparty / OpenAI concentration. A single un-IPO'd customer ($60B/yr, 2027–2031) is the dominant chunk of a backlog of which the 10-K confirms only 12% converts within a year. If OpenAI under-consumes/renegotiates, Oracle simultaneously loses the revenue, owns purpose-built assets worth less, and may write down — and the financing is circular (the same capital ecosystem funds OpenAI and its suppliers).
2. The funding wall — now audited. $129.5B debt, $26.6B operating-lease liabilities, a NEW $19B post-quarter purchase commitment, FY27 capex ~$70B, FY26 FCF −$23.7B, three agencies on negative/cautious outlook, leverage trending toward the ~4x trigger, and a full year of 6.5% preferred dividends ahead. Oracle is funding a widening multi-year cash deficit into a window where any credit-appetite softening bites.
3. Margin / commoditization — now visible in the numbers. The cloud-and-software segment margin already slipped 63%→59% as infra expense (+56%) outran revenue (+19%). GPU rental is thin, capital-intensive, NVIDIA-gated, with no Oracle silicon edge vs Google/Amazon. As supply normalizes and the customer-supplied-GPU trend grows, OCI economics compress exactly as depreciation and interest peak.

Pre-mortem (18 months out, thesis broke). Late 2027: OpenAI's growth disappointed, it renegotiated its Oracle commitment down, one Stargate site sits underutilized. Oracle takes an impairment, RPO "growth" reverses for the first time, the securities class action survives the motion to dismiss (briefed Dec 2026, ruled 2027), a downgrade lifts the cost of the remaining build, and the 13% workforce cut proves to have been the start not the end. The stock that already round-tripped to $166 breaks $120. Post-mortem line: the backlog was real, but it was a loan to one borrower, and the borrower's cash flows didn't show up on time.

Are multiples too high? At ~31.6x trailing / ~22x forward, you pay a (de-rated but still) software multiple for a business with the cash-flow and leverage profile of a leveraged infrastructure developer. On EV/FCF it is negative. Defensible only if you believe the FY29–30 waterfall; on audited current cash flow it is not.

Contrarian view (what the market refuses to see). The bears are right about the risk but may be wrong about the price: the stock has already de-rated ~52% and now prices a meaningful probability of an OpenAI disappointment plus an operating-stress narrative (the layoff selloff). The genuinely contrarian read is that the market over-corrected on counterparty fear while under-appreciating (a) the diversification of the backlog beyond OpenAI (xAI, Meta, federal, enterprise multicloud), (b) the margin tailwind from a 13% lower opex base, and (c) the optionality of the database-AI-agent attach. The risk is real; the question is whether −52% and a ~22x forward already pay you for it.

Lens 13 · Devil's Advocate (short-seller) — re-run

Dismantling the bull case. Oracle converted a high-margin, cash-gushing software annuity into a low-margin, capital-devouring, single-counterparty infra bet financed with $129.5B of debt + $5B preferred + $26.6B leases + a fresh $19B purchase commitment. The audited 12%/34%/34%/remainder RPO curve is the short thesis on a plate: ~88% of the celebrated $638B is dated beyond a year, ~32% beyond five — Oracle is being valued on a backlog whose payment is mostly a 2028–2031 event from a borrower (OpenAI) that isn't yet public and isn't yet cash-generative at the required scale. The dangerous concentration is in the growth and the backlog, not the legacy book. The "moat" is willingness to lever, not technology — and the 63%→59% segment-margin slip already shows commoditization biting. The most dangerous competitor bulls underestimate isn't AWS — it's the customers vertically integrating (the customer-supplied-GPU contracts literally reveal the trend) plus the neoclouds, compressing Oracle from landlord to operator. Worst capital allocation: abandoning the buyback model for a bet-the-balance-sheet pivot, moving the CFO-minded CEO sideways right before FCF went negative, reaching for a mandatory convertible and committing another $19B post-quarter — and now cutting 13% of staff, which a bull calls discipline and a bear calls the first crack. Assumptions that must hold for ~$166: OpenAI + the whales consume on schedule for five years; credit stays accommodating through a widening deficit; NVIDIA allocation and power arrive on time; OCI margin stops falling; the securities suit fails; the FY29–30 waterfall materializes. If growth disappoints 20–30%: EPS bear ($6.50–$7.20) + a de-rate toward an infrastructure multiple implies real downside even from $166. Single permanent-impairment scenario: OpenAI materially renegotiates/fails to fund mid-build while Oracle is over-levered → impairment + downgrade + RPO reversal + the class action gaining traction. Plausibility: moderate and rising — which is precisely why the stock more than halved while results beat.

Lens 14 · Management Questions — carried forward (15, unchanged — still the right questions)

Carried from deep-dive-2026-06-18.md; the FY26 10-K answered none of them and made #1–#4 more urgent. Ordered by information value:

1. Of the $638B RPO, what fraction is attributable to OpenAI/Stargate, and what are the take-or-pay/minimum-commitment and termination terms if the customer under-consumes?
2. What contractual/credit protection exists if OpenAI cannot fund — prepayments held, GPU ownership, parent guarantees, letters of credit?
3. Walk us through the precise path to positive FCF: which fiscal year, at what capex run-rate, and what RPO-conversion assumption gets you there? (The audited 12%-in-12-months curve makes this sharper.)
4. You said most incremental RPO needs no incremental Oracle funding — yet you committed a new $19B post-quarter and guide ~$70B FY27 capex. Reconcile with a sources-and-uses bridge through FY28.
5. What is OCI's current gross margin at today's utilization (segment margin slipped to 59%), and where does it settle vs the Big Three given no proprietary silicon?
6. What economic life/depreciation schedule applies to the AI GPU estate, and what impairment exposure if a major contract is renegotiated?
7. How diversified is the backlog beyond OpenAI — what share is xAI, Meta, federal, enterprise multicloud, and how concentrated are the top three?
8. At what S&P/Moody's leverage threshold do you slow the build, and how do you protect the rating through the FY27–28 deficit?
9. How much of FY27's ~$70B capex is GPUs (NVIDIA-gated) vs power/shell, and where are the binding constraints?
10. Size, dilution, and trigger of the Mandatory Convertible Preferred — should investors expect further equity issuance?
11. On customer-supplied-GPU contracts, how is revenue recognized, and how does the margin/asset-ownership differ vs Oracle-owned-GPU deals?
12. What is your response to the securities class action's core allegation about the cloud disclosures (amended complaint due July 2026), and what guidance-practice changes follow?
13. Why move Safra Catz to Executive Vice Chair precisely as the company entered its largest-ever capital-deployment phase — who owns financial discipline now?
14. With a 13% workforce cut (Revenue/Health Sciences ~−30%), is this a one-time efficiency reset or an ongoing program, and what does the Health Sciences cut imply for the Cerner growth case and goodwill?
15. What is the smallest OpenAI-consumption shortfall that would force a guidance cut or impairment, and how are you stress-testing for it?

---