This is Part 2 of a six-part investigative series analyzing Oracle’s $300 billion OpenAI partnership and the financial risks it creates. Each article examines a different dimension of what may be the most consequential bet in enterprise technology history.
Part 1 explored the historical parallels between Oracle’s current situation and its 1990 near-bankruptcy crisis, establishing the framework for understanding how a company can face existential risk twice, 35 years apart.
Part 1: Oracle’s $315 Billion Reality Check – When History Rhymes with 1990
Part 2 (below) dissects the deal mechanics that Wall Street initially celebrated, then violently rejected—examining why revenue recognition timing, collection probability, and customer concentration create a cascade of compounding risks.
Coming in this series:
- Part 3: Oracle’s RPO Shell Game: How $455 Billion in contracted revenue became a red flag
- Part 4: The 10% Layoff Signal: Oracle’s most reliable crisis indicator across 35 years
- Part 5: The Credit Crisis: When bond markets scream danger and CDS spreads hit multi-year highs
- Part 6: The Oracle Endgame: A synthesis of all risk factors and what they mean for stakeholders
Each analysis stands alone but builds on previous work. You can read them in any order, though the full picture emerges when taken together. I also want to thank those of you who commented and pushed back on some of my arguments. I think open and respectful dialog is great and I think the discourse has improved the quality of Part 2 from the original draft. Also…
The market answered my question faster than I expected.
Inside Oracle’s OpenAI Deal – Why Wall Street Lost Faith
When I published Part 1 on Friday morning, the question was whether Oracle’s credit risks were fully priced in. By 4:00 PM, the market offered a definitive “no.”
Oracle shares plunged 5.7% to close at $198.76, shattering the psychological $200 barrier. This single-day sell-off wiped out an additional $34 billion in shareholder value, bringing total destruction since September to nearly $400 billion. The catalyst? Barclays officially downgraded Oracle’s corporate debt to “Underweight,” citing the exact “weakening balance sheet” risks I highlighted.
The $300 billion OpenAI partnership Oracle announced in September represents one of the most audacious bets in tech history. The market’s violent rejection suggests investors have reached similar conclusions about execution risk.
The Deal Mechanics Nobody Wants to Discuss
Oracle’s partnership centers on the “Stargate” initiative, a $500 billion infrastructure platform to achieve 10 gigawatts of AI data center capacity.¹ Within this framework, Oracle committed to develop 4.5 gigawatts of new capacity, bringing the partnership’s total to over 5 gigawatts. The payment structure reveals the fatal flaw: Cash generation and revenue recognition are effectively deferred until 2027, when OpenAI must deliver approximately $60 billion annually through 2032.²
This creates an unprecedented cash flow timing mismatch. Oracle must spend $35 billion in fiscal 2026 alone on capital expenditures (a 270% increase from the prior year) while revenue recognition doesn’t begin for two years.³ Moody’s forecasts Oracle will have negative free cash flow of “$6 billion or worse” in fiscal 2026, with Fitch projecting “negative pre-dividend FCF exceeding $11 billion” in fiscal years 2026 and 2027.⁴
Why Revenue Recognition Doesn’t Begin Until 2027: The ASC 606 Problem
The two-year delay before revenue recognition isn’t arbitrary. It’s driven by fundamental accounting requirements under ASC 606 that Oracle cannot circumvent.
Performance obligations must be satisfied before revenue recognition. Under ASC 606-10-25-27, revenue can only be recognized over time when specific criteria are met, including when “the customer simultaneously receives and consumes the benefits as the entity performs.”⁵ For Oracle’s infrastructure buildout, this means revenue recognition cannot begin until the data centers are operational and OpenAI is actually consuming computing capacity.
The timing creates a critical vulnerability. Oracle incurs massive capital expenditures in 2025-2027 to construct facilities, procure GPUs, and build interconnect infrastructure. But under ASC 606-10-25-25, “control” (defined as “the ability to direct the use of, and obtain substantially all of the remaining benefits from, the asset”) doesn’t transfer to OpenAI until facilities are complete and computing capacity is available.⁶
This isn’t a choice. It’s an accounting requirement. Oracle cannot recognize revenue for infrastructure it hasn’t yet delivered, regardless of contractual commitments. Unless the contract contains distinct ‘build-to-suit’ leasing components that allow for earlier recognition (which is rare for standard cloud consumption contracts like this), the company must finance 2-3 years of negative cash flow before any revenue materializes.
Construction delays, permit issues, and equipment supply chain problems routinely extend infrastructure projects by 6-12 months. If Oracle’s data centers come online in late 2027 or early 2028 rather than mid-2027, the negative cash flow period extends accordingly. And OpenAI won’t pay for capacity that isn’t operational.
The Collection Probability Question Revisited
ASC 606-10-25-1(e) requires that “it is probable that the entity will collect substantially all of the consideration to which it will be entitled” for a contract to exist for accounting purposes.⁷ The term “probable” in U.S. GAAP is generally accepted to mean 75-80% likelihood.⁸
Here’s the uncomfortable question: Can Oracle legitimately assert 75-80% confidence that OpenAI will pay $60 billion annually starting in 2027?
Consider what must happen for OpenAI to meet its obligations:
- Revenue growth from $12B to $60B+ (5x increase in two years)
- Simultaneous profitability achievement while currently losing $9-14 billion annually
- Sustained AI demand at sufficient scale to justify massive capacity purchases
- Access to capital markets to bridge the gap until profitability (potentially $100B+ in funding)
- No competitive displacement by Microsoft, Google, or Anthropic
Each of these represents substantial execution risk. Collectively, they must all succeed for Oracle to collect substantially all consideration. Does this probability distribution genuinely exceed 75-80%?
If Oracle applied the collection probability threshold conservatively (as ASC 606 requires when uncertainty is high), a material portion of the $300 billion OpenAI contract might not meet the criteria for inclusion in RPO at all. The fact that Oracle included the full amount suggests either extraordinary optimism about OpenAI’s prospects or aggressive interpretation of accounting standards.
OpenAI’s Financial Reality: The Elephant in the Data Center
OpenAI’s financials are sobering. The company generated $3.7 billion in revenue in 2024 while posting a $5 billion loss. For 2025, projections show approximately $12 billion in revenue against $9 billion in losses.⁹
OpenAI’s cost structure is dominated by inference expenses. In the first half of 2025 alone, OpenAI spent $5.02 billion on inference with Microsoft Azure, with cumulative costs reaching $8.67 billion by September.¹⁰ The company expects cumulative cash burn of $115 billion through 2029, with operating losses potentially reaching $74 billion in 2028 alone.¹¹
The arithmetic is unforgiving. For OpenAI to meet its Oracle obligations while continuing operations, it must secure substantial external funding. If OpenAI generates $20 billion in revenue in 2027 but must pay Oracle $60 billion while covering $10+ billion in operational expenses, the funding gap exceeds $50 billion annually.
This isn’t theoretical. OpenAI must actually raise and deploy this capital. If AI investment sentiment cools, if venture capital becomes scarce, or if OpenAI’s competitive position weakens, the company may be unable to meet its Oracle commitments regardless of contractual obligations.
The Infrastructure Gamble: Building for an Uncertain Future
Oracle is developing 4.5 gigawatts of additional capacity across Texas, New Mexico, Wisconsin, and Michigan.¹² The flagship Stargate I site in Abilene, Texas, houses over 400,000 GPUs with 1.4 gigawatts of power.
These aren’t just data centers. They’re specialized, hyperscale facilities designed for specific AI workloads. Specialized systems create significant lock-in risk. If OpenAI fails, repurposing these facilities could be extraordinarily difficult. Scott Bickley of Info-Tech Research Group identified the “Black Swan scenario”: Oracle could be left with significant “stranded infrastructure and capital.”¹³
The depreciation challenge compounds this risk. Michael Burry (who predicted the 2008 financial crisis) accused Oracle of overstating earnings by extending equipment useful life to 6 years, when AI hardware has 2-3 year obsolescence cycles.¹⁴ If Burry is correct, Oracle’s earnings are significantly overstated, and the actual cost of replacing obsolete equipment will far exceed depreciation charges.
The accounting implications are serious. If Oracle’s depreciation schedules understate the true rate of GPU obsolescence, the company will face massive write-downs when equipment must be replaced sooner than expected. A 6-year depreciation schedule on assets with 2-3 year useful lives creates a cumulative gap that compounds annually. By 2028-2029, Oracle could face tens of billions in asset impairment charges if Burry’s analysis is correct.
Why Credit Markets Are Panicking
Oracle executed an $18 billion bond sale in September 2025 (the market’s second-largest investment-grade deal of the year) explicitly for AI infrastructure investments.¹⁵ Despite this substantial issuance, Morgan Stanley estimates Oracle needs an additional $55-75 billion in debt offerings.¹⁶
The market reaction has been brutal. A $3.5 billion, 30-year Oracle bond dropped approximately 8% since its October peak, trading at 65 cents on the dollar.¹⁷ A 35% discount indicates investors anticipate refinancing pressure or default risk.
All three major credit rating agencies maintain Oracle at investment-grade levels, but Moody’s and S&P have negative outlooks. Moody’s explicitly flagged “significant counterparty risk” due to Oracle’s reliance on OpenAI, warning the contract could lead to “an extended period of high leverage and negative cash flow.”¹⁸
Credit markets understand the timing mismatch better than equity investors initially did. Bond buyers are lending Oracle money today that won’t generate returns until 2028-2030 (assuming everything goes perfectly). If anything goes wrong (construction delays, OpenAI funding challenges, AI market correction), Oracle faces years of negative cash flow with no revenue to offset interest expense.
The Concentration Risk Nobody Calculated
Customer concentration above 20% raises red flags in enterprise technology. Oracle’s 66% RPO concentration with OpenAI is unprecedented among major technology companies. Historical precedents are sobering:
- Nortel Networks concentrated revenue with telecom customers who couldn’t pay, leading to bankruptcy
- Enron concentrated trading relationships that evaporated, resulting in bankruptcy
- Lehman Brothers concentrated exposures to real estate, ending in bankruptcy
When D.A. Davidson stated investors should “definitely keep everything they say from now on with a little more grain of salt,” it reflected the credibility damage from decades of overpromising.¹⁹
The Revenue Recognition Risk Cascade
The combination of collection uncertainty, revenue recognition timing delays, and customer concentration creates a cascade of compounding risks:
If OpenAI experiences funding difficulties in 2026: Oracle continues spending $35-60 billion annually on infrastructure for a customer that may not be able to pay. Revenue recognition remains zero because facilities aren’t operational. Oracle’s debt load continues increasing with no offsetting revenue.
If data center construction delays occur: Revenue recognition gets pushed from 2027 to 2028 or later. Oracle’s negative cash flow period extends by 12-18 months. Credit rating agencies downgrade Oracle to junk status, triggering covenant violations and refinancing crises.
If AI market demand disappoints: OpenAI renegotiates payment terms (converting the “firm” $300 billion commitment to the “flexible arrangement” Jim Chanos suggests it actually is). Oracle recognizes only a fraction of projected revenue, leaving the company overleveraged with underutilized assets.
Under ASC 606-10-32-14, entities must “update the estimated transaction price (including updating its assessment of whether an estimate of variable consideration is constrained) at the end of each reporting period.”²⁰ If facts and circumstances change, Oracle may be required to reduce its RPO, adjust revenue projections, and recognize impairment losses on infrastructure investments that no longer have identifiable revenue streams.
The Verdict
Oracle’s OpenAI partnership represents a $300 billion bet on exponential AI growth, OpenAI’s unprecedented revenue scaling, perfect execution of massive infrastructure buildout, and sustained access to debt markets at reasonable rates. Any single failure point could trigger catastrophic consequences.
The market has spoken: Nearly $400 billion in destroyed market value, credit default swaps at multi-year highs, and analyst downgrades across the board. Oracle is essentially betting the company on forces entirely beyond its control.
The revenue recognition timing alone (2-3 years of massive negative cash flow before any revenue materializes) would concern prudent financial managers. Add customer concentration of 66%, collection probability questions with an unprofitable counterparty, and aggressive depreciation schedules, and you have a situation where multiple risk factors compound into potential catastrophe.
Analyzing deals of this magnitude requires examining both the financial mechanics and the execution realities. This is the kind of deal that looks transformative in PowerPoint presentations but faces brutal reality when execution begins. Oracle is discovering that reality now, and credit markets are pricing in the possibility that transformation becomes bankruptcy.
Bibliography
- OpenAI: “Stargate Advances with Partnership with Oracle”
- Oracle Investor Relations: “Q1FY26 Earnings Release” (September 9, 2025)
- CRN: “Oracle Q1 Takeaways For The Channel”
- Investing.com: “Moody’s revises Oracle’s outlook to negative” / Fitch Ratings
- Financial Accounting Standards Board (FASB). “ASC 606-10-25-27: Revenue Recognition Over Time – Criteria.” Accounting Standards Codification.
- Financial Accounting Standards Board (FASB). “ASC 606-10-25-25: Control Transfer Definition.” Accounting Standards Codification.
- Financial Accounting Standards Board (FASB). “ASC 606-10-25-1(e): Contract Criteria – Collection Probability.” Accounting Standards Codification.
- PWC IFRS and US GAPP: Similarities and differences guide
- LessWrong: “OpenAI lost $5 billion in 2024 and its losses are accelerating”
- Where’s Your Ed At: “OpenAI Spends On Inference and Revenue Share With Microsoft”
- Fortune: “OpenAI says it plans to report stunning annual losses through 2028”
- Data Center Frontier: “OpenAI and Oracle’s $300B Stargate Deal”
- CIO: “Oracle’s cloud strategy: An increasingly risky bet”
- Fortune: “The ‘Big Short’ investor betting $1 billion against the AI bubble”
- Bloomberg: “Oracle looks to raise $15 billion from corporate bond sale”
- The Register: “Oracle will have to borrow at least $25B a year”
- Axios: “Big Tech’s debt problem” (November 17, 2025)
- Investing.com: “Moody’s revises Oracle’s outlook to negative”
- Benzinga: “Top Analyst Says Oracle’s ‘Irresponsible’ AI Bet”
- Financial Accounting Standards Board (FASB). “ASC 606-10-32-14: Updates to Transaction Price Estimates.” Accounting Standards Codification.