Oracle (ORCL): AI Backlog, Capex, Credit & the Best and Worst Cases
Executive assessment
Oracle is two companies sharing one balance sheet. The first is a mature database, applications and support business that throws off dependable cash. The second is an AI infrastructure builder that is spending roughly one and a half dollars on data centers for every dollar of total company revenue. Demand for the second business is not the problem: contracted backlog reached $664 billion at the end of August, and cloud infrastructure revenue grew 121% last quarter. The problem is timing. The money goes out years before the contracted revenue comes in, and roughly half of that backlog is widely estimated to depend on a single customer, OpenAI, which itself burns cash.
The credit market has stopped giving Oracle the benefit of the doubt. Five-year credit default swaps more than doubled this year to a record of about 315 basis points on October 19, a level that a market analyst at S&P Global Market Intelligence said is more in line with high-yield credit. S&P already rates Oracle BBB-, the last rung of investment grade. The stock, at $129, is down about 55% from its 52-week high.
Our scenario work below comes to a simple conclusion: Oracle does not have a demand problem, it has a delivery-and-funding problem, and almost everything turns on whether backlog converts on Oracle's stated schedule through fiscal 2029. If it does, today's price looks deeply discounted. If OpenAI or data-center power stalls, leverage pushes Oracle into junk territory and equity holders absorb most of the damage.
What changed since our October 20 version
- First-quarter fiscal 2027 results (quarter ended August 31) are now incorporated, including the 10-Q balance sheet, lease commitments and the backlog conversion schedule.
- The $664B backlog figure is now verified against Oracle's own release and 10-Q. It was flagged as unverified in the earlier version.
- The "12% converts within 12 months" claim is now resolved. Oracle's CFO gave 12% of the $638B backlog in June; the August 10-Q gives 13% of $664B. The OpenAI concentration figure remains an analyst and rating-agency estimate, not an Oracle disclosure.
- New events: Oracle's force majeure notice on Project Jupiter (September 24), the October 14 AI-infrastructure selloff, and the move of Oracle's CDS to records near 288–315bp.
- New section: best, base and worst-case scenarios through fiscal 2029.
Latest quarter: Q1 fiscal 2027
| Three months ended Aug 31 | 2025 | 2026 | Change |
|---|---|---|---|
| Total revenue | $14.93B | $19.35B | +30% |
| Cloud infrastructure (OCI) revenue | $3.35B | $7.39B | +121% |
| Cloud applications (SaaS) revenue | $3.84B | $4.22B | +10% |
| Software license + support revenue | $5.72B | $5.55B | −3% |
| GAAP operating income | $4.28B | $6.73B | +57% |
| Interest expense | $0.92B | $1.43B | +55% |
| Depreciation | $1.35B | $3.16B | +134% |
| Operating cash flow | $8.14B | $23.10B | +184% |
| of which: customer prepayments with financing component | — | $11.36B | |
| Capital expenditure | $8.50B | $28.50B | +235% |
| Free cash flow (reported) | −$0.36B | −$5.40B | |
| Free cash flow excluding customer prepayments | −$0.36B | −$16.76B |
Source: Oracle Form 10-Q for the quarter ended August 31, 2026, and Q1 FY27 earnings release (Sept. 10, 2026). The prepayment-adjusted line is our calculation.
The headline "record operating cash flow" needs one adjustment. Nearly half of it, $11.4 billion, was customers paying Oracle in advance for capacity, and Oracle's 10-Q classifies those payments as containing a significant financing component. That money is real and it reduces Oracle's own borrowing needs, but it is not recurring operating earnings. Strip it out and the quarter burned about $17 billion before financing. Oracle funded the gap with a $19.9 billion at-the-market stock sale that added 141 million shares, about 4.9% of the share count. It issued no new bonds and repaid $4.2 billion of debt.
Operating trends worth noting
- Oracle delivered 850MW of data-center capacity and more than 300,000 GPUs to AI customers during the quarter, roughly triple the prior quarter's pace.
- Management reported AI infrastructure utilization near 98% and higher prices on GPUs renewed or resold, which supports the claim that demand exceeds supply.
- The legacy software line is shrinking slowly (−3%) as customers move to cloud. Software support at $4.9B a quarter is still the anchor of Oracle's cash generation.
- Guidance: Q2 revenue growth of 30–34%; fiscal 2027 revenue of at least $90 billion and non-GAAP EPS of $8.10. Management has guided to about $70 billion of its own capital spending in fiscal 2027, plus $20–25 billion more it expects customers to reimburse, without a timeline for those repayments.
Cash flow history
| Fiscal year ended May 31 | 2024 | 2025 | 2026 | TTM Aug 2026 |
|---|---|---|---|---|
| Operating cash flow | $18.7B | $20.8B | $32.0B | $46.9B |
| Capital expenditure | $6.9B | $21.2B | $55.7B | $75.7B |
| Free cash flow | $11.8B | −$0.4B | −$23.7B | −$28.7B |
FY2025–2026 from Oracle's FY2026 Form 10-K; FY2024 rounded historical comparatives. Trailing twelve months computed as FY2026 minus Q1 FY2026 plus Q1 FY2027 from the 10-Q. TTM operating cash flow includes $11.4B of customer prepayments.
The backlog: large, real, and slow
Remaining performance obligations (RPO) are signed contracts for work Oracle has not yet delivered. They are not cash and not guaranteed profit. Oracle's 10-Q gives the expected recognition schedule for the $664 billion:
| Recognition window (from Aug 31, 2026) | Share | Dollars | Per year |
|---|---|---|---|
| Next 12 months | ~13% | ~$86B | ~$86B |
| Months 13–36 | ~37% | ~$246B | ~$123B |
| Months 37–60 | ~34% | ~$226B | ~$113B |
| Thereafter | ~16% | ~$106B |
Percentages from the Q1 FY27 Form 10-Q; dollar amounts are our arithmetic on $664B.
Two things stand out. First, the heavy years are fiscal 2028 and 2029, which is exactly when the $288 billion of not-yet-started data-center leases begin to run. If the backlog recognizes on schedule, revenue in those years should roughly double from fiscal 2026's $67 billion. Second, every year of delay compounds: Oracle pays for construction and leases on a fixed calendar, while backlog revenue only starts when capacity is live.
Customer concentration
Oracle has never named OpenAI or put a dollar figure on its OpenAI contracts in an SEC filing. The widely reported contract is about $300 billion over roughly five years. When S&P downgraded Oracle in July, it reportedly estimated that OpenAI accounts for roughly half of RPO, and Bank of America analysts have put it above 50%. We treat "about half" as a credible estimate rather than a fact.
That matters because OpenAI's own finances are stretched. According to a Financial Times report in September, an internal OpenAI presentation projects about $278 billion of negative free cash flow from 2026 through 2030, with revenue rising from roughly $36 billion this year to $350 billion in 2030. OpenAI raised $122 billion in March at an $852 billion valuation. Its ability to keep paying Oracle depends on continued access to private capital, and that access is the single largest variable in Oracle's credit story.
Oracle's partial defense is structure. Most of the backlog added in the last three quarters came with customer prepayments or customer-supplied GPUs, and Oracle said the $30 billion of contracts booked in Q1 add nothing to its capital-raising plans. A customer that has prepaid has some skin in the game; one that has supplied its own GPUs leaves Oracle with less hardware to strand.
The full balance sheet, including what is not on it
| As of August 31, 2026 | Amount | Note |
|---|---|---|
| Notes payable and borrowings | $125.3B | $7.6B current. Fair value was $105.7B, about 85% of carrying value, reflecting wider spreads and higher rates. |
| Cash and marketable securities | $37.1B | Plus $2.6B restricted cash. |
| Net debt | ~$88B | Our calculation; excludes leases. |
| Operating + finance lease liabilities | $43.8B | On balance sheet. |
| Leases signed but not commenced | $288B | Mostly data centers; start between Q2 FY27 and FY29; terms of 15–19 years. Not on the balance sheet. |
| Unconditional purchase obligations | $34.2B | Up from $13.3B in May; mostly components and power. |
| Deferred revenue (customer prepayments) | $30.8B | Doubled from $15.4B in May. |
| Shareholders' equity | $67.2B | Up from $43.1B, mainly from the $19.9B stock sale. |
Source: Oracle Q1 FY27 Form 10-Q.
The $288 billion lease figure is the most important number most investors overlook. Leases of 15–19 years are being signed to serve cloud contracts that, per Oracle's 10-Q, generally run one to five years. If customers renew, the long leases are an asset; if they don't, Oracle owns a decade or more of rent on buildings it may struggle to refill. Rating agencies add leases to debt, which is why S&P measures Oracle's adjusted leverage in the mid-4x range even though borrowings alone are about 3x EBITDA.
What the credit market is saying
- Ratings. S&P cut Oracle to BBB- (stable) on July 9, citing fiscal 2027 capex of $90–95 billion, a free operating cash flow deficit of about $42 billion and adjusted leverage in the mid-4x range. S&P has said it could cut again if it expects leverage to stay above 4.5x or if Oracle does not look on track for positive free operating cash flow by fiscal 2029. Moody's rates Oracle Baa2 with a negative outlook, one notch higher; Fitch rates it BBB.
- What 300bp implies. A rough conversion of a ~290–315bp spread, assuming 40% recovery, implies a market-priced chance of default of about 21–23% over five years. That is a risk-neutral price that includes a premium for uncertainty and hedging demand. It overstates the real-world probability, but the direction is clear: in January the same arithmetic gave about 11%.
- Why it matters for equity holders. A cut to junk by two agencies would force some investment-grade-only funds to sell, raise Oracle's cost of new money, and make the planned debt-plus-equity funding mix more expensive. Oracle has said preserving investment grade is its top financial priority and has pledged no new bonds in calendar 2026.
Execution risk: Project Jupiter
On September 24, Oracle sent a force majeure notice to the Blue Owl-owned developer of Project Jupiter, a Stargate campus in southern New Mexico. The notice preserves Oracle's right to defer full rent, reportedly for up to three years, if the campus misses its 2028 start because power is not ready. A gas pipeline to the site has slipped to February 1, 2027, after permit denials, and the state's decision on the air permit for the site's fuel-cell power system is due by November 23. Oracle says the project remains on schedule. About 20 banks provided an $18 billion construction loan that has been reported trading at stressed levels.
Force majeure protects Oracle's cash in the short run. The concern is what it signals: power, not chips or demand, has become the binding constraint, and any campus that slips pushes backlog recognition into later years while lease and construction commitments stay on the calendar.
Scenarios: best, base and worst case through fiscal 2029
We built a simple three-year model starting from Oracle's May 31, 2026 balance sheet. It is meant to show how sensitive Oracle's credit and equity are to backlog conversion, not to predict outcomes.
Assumptions
| Driver | Best case | Base case | Worst case |
|---|---|---|---|
| Story | Backlog converts on Oracle's 10-Q schedule; power arrives; OpenAI keeps raising capital | Conversion slips 6–12 months; Jupiter delayed; margins compress as GPU mix grows | OpenAI cuts or renegotiates commitments; multiple campuses delayed; Oracle slows capex |
| Revenue FY27 / FY28 / FY29 | $93B / $138B / $175B | $90B / $120B / $148B | $86B / $100B / $106B |
| EBITDA margin | 48% throughout | 47% → 44% | 46% → 38% |
| Net capex (after customer funding) | $70B / $70B / $55B | $70B / $80B / $70B | $78B / $65B / $40B |
| Funding mix for any shortfall | 40% debt / 60% equity | 50 / 50 | 80 / 20 (equity window shut) |
| Valuation multiple (EV / FY29 EBITDA) | 14x (12–16x) | 11x (9–13x) | 7x (5–9x) |
Common assumptions: 5.2% average cost on gross debt; 15% cash tax rate; $2.00 annual dividend per share maintained; minimum cash balance of $25B; the $19.9B Q1 stock sale included in FY27. Leases are treated as operating costs inside EBITDA, so the leverage shown is borrowings to EBITDA, lower than agency lease-adjusted figures. Model is ours and illustrative.
Results
| Best | Base | Worst | |
|---|---|---|---|
| Free cash flow FY27 | −$35B | −$37B | −$48B |
| Free cash flow FY28 | −$16B | −$36B | −$32B |
| Free cash flow FY29 | +$15B | −$17B | −$9B |
| Cumulative FCF FY27–29 | −$36B | −$91B | −$89B |
| New debt raised FY27–29 | $15B | $41B | $64B |
| Gross borrowings, end FY29 | $144B | $171B | $194B |
| Borrowings / EBITDA, FY29 | 1.7x | 2.6x | 4.8x |
| Share count increase vs May 2026 | +9% | +15% | +12% |
| Illustrative value per share (mid / range) | $336 ($282–389) | $180 ($139–220) | $40 ($14–65) |
| vs. $129 close, Oct 21 | +160% | +39% | −69% |
Illustrative value = multiple × FY29 EBITDA minus FY28 net debt, divided by FY28 share count; a rough two-year-forward view, not a price target. For reference, the Street's average 12-month target was about $237 across 43 analysts as of October 20, with a range of $110–$400. Our base case FY27 free cash flow (−$37B) sits between the consensus figure cited by CNBC (−$36B) and S&P's estimate (about −$42B).
Best case
The backlog converts on schedule and the market re-rates Oracle as a profitable hyperscaler.
- Revenue roughly doubles from FY26 by FY28 as the 37% tranche of RPO recognizes.
- FCF turns positive in FY29, inside S&P's deadline; leverage falls below 2x borrowings/EBITDA.
- CDS retraces toward 100–150bp; S&P outlook stabilizes and Moody's removes its negative outlook.
- Main cost to shareholders: about 9% dilution, mostly already done.
Needs: OpenAI funded through 2028, Jupiter and other campuses energized on time, GPU pricing holding up.
Base case
Demand is real but delivery runs late, so Oracle stays a funding story for another two years.
- FCF stays negative through FY29, missing S&P's positive-FCF test by a year or more.
- About $90B of cumulative cash burn funded half with debt and half with stock; dilution about 15%.
- Likely one more agency action: a Moody's cut to Baa3 is plausible. A cut to junk by S&P is a real risk but not the central case.
- Equity works if the market looks through to FY30 earnings; it stalls if it doesn't.
Signals: RPO conversion near but below schedule, capex guidance raised again for FY28.
Worst case
The anchor customer pulls back and Oracle is left with long leases and short contracts.
- OpenAI fails to raise enough, renegotiates or defers. RPO is cut and revenue growth stalls near $100B.
- Capex already committed keeps cash burn high in FY27–28; Oracle borrows about $64B more at worse rates.
- Borrowings near 5x EBITDA and lease-adjusted leverage well above 5x. Fallen-angel downgrade to high yield by at least two agencies; CDS above 500bp.
- Dividend and buyback likely suspended; asset write-downs on stranded capacity.
Key point: equity takes most of the damage. Even here, the software support business keeps generating cash, so a near-term default is unlikely; the risk is a long, expensive deleveraging.
Why the range is so wide
Leverage magnifies small changes. Moving FY29 EBITDA by $10 billion moves the illustrative equity value by roughly $35 per share at an 11x multiple, because net debt is fixed while the multiple applies to earnings. The same mechanism that makes the best case look cheap makes the worst case look severe. This asymmetry is what the credit market is pricing.
Signposts to watch
| Signal | Bullish reading | Bearish reading |
|---|---|---|
| New Mexico air-permit decision for Jupiter (due by Nov 23, 2026) | Permit granted; force majeure not used | Denial or delay; force majeure invoked |
| Q2 FY27 results (expected December) | OCI growth ≥120%; FCF ex-prepayments improving; RPO conversion on schedule | Higher FY27 capex; prepayments fading; RPO 12-month share falling |
| Ratings | Moody's outlook to stable | Moody's to Baa3, or any agency to negative watch |
| 5-year CDS | Back below 200bp | Sustained above 400bp |
| OpenAI financing | New round closes at or above $1T valuation; IPO filing | Round delayed or down; compute commitments cut further |
| Fiscal 2028 capex guide (likely June 2027) | Flat to down with customer funding rising | Up again without matching prepayments |
Bottom line
Oracle's backlog is real and its software business is a genuine cushion, but the company has committed to costs on a fixed schedule in exchange for revenue on a contingent one. At $129, the stock prices in a meaningful chance that the contingent part slips. Whether that is an opportunity or a trap depends less on AI demand than on two narrower questions: does the power arrive on time, and can OpenAI keep paying?
Method and limitations
- The OpenAI share of RPO (about half) is a rating-agency and analyst estimate; Oracle has not disclosed it.
- CDS readings come from multiple data providers via press reports and can differ by several basis points on the same day.
- The scenario model is deliberately simple: it ignores working-capital swings, changes in prepayment timing, asset sales, and the effect of new leases commencing. Lease commencement would raise agency leverage in every scenario.
- Valuation ranges are an illustration of sensitivity, not price targets.
Sources
- Oracle Form 10-Q, quarter ended August 31, 2026 (SEC)
- Oracle Q1 FY27 earnings release, Sept. 10, 2026
- Oracle FY2026 Form 10-K
- Reuters via Outlook Business: FY27 capex and financing plan (June 2026)
- Wealth Professional: S&P downgrade triggers, Moody's affirmation
- Reuters via Euronext: Oracle ratings analysis
- CNBC, Oct. 19, 2026: Oracle CDS at 315bp
- CNBC Daily Open, Oct. 20, 2026
- Securitas Global: CDS record 288bp, Oct. 15
- CDS at ~203bp after S&P downgrade (July 2026)
- TechCrunch: Project Jupiter force majeure
- Capacity: what the force majeure notice does
- SiliconANGLE: Oct. 14 AI-infrastructure selloff
- Reuters/FT via DealStreetAsia: OpenAI $278B cash burn forecast
- BofA estimate of OpenAI share of RPO (secondary report)
- StockAnalysis: analyst targets (Oct. 20, 2026)
- Morningstar: ORCL quote, Oct. 21, 2026
Disclaimer: Personal investment views for educational purposes, not individualized investment, financial or tax advice. All investments involve risk, including loss of principal. Consult with your financial advisor before making any decisions based on this educational article. Figures are historical snapshots, not live quotes. Scenario outputs are hypothetical.