Oracle Credit Downgrade In-Depth Analysis: Can $638 Billion RPO Convert to Cash Flow Per Share

On July 9, 2026, S&P Global Ratings downgraded Oracle's long-term issuer credit rating from BBB to BBB-, and downgraded its short-term rating from A-2 to A-3, with a stable outlook. BBB- remains investment grade, but is already the lowest tier within S&P's investment grade ratings.

The rating adjustment occurred during a phase of significant acceleration in Oracle's operational growth. In fiscal year 2026, Oracle's total revenue reached $67.4 billion, up 17% year-over-year; OCI infrastructure revenue reached $18.1 billion, up 77% year-over-year; as of May 31, 2026, the company-wide remaining performance obligations, or RPO, reached $638 billion, up 363% year-over-year. Oracle stated that the majority of the RPO increase in the third and fourth fiscal quarters came from large AI contracts.

Therefore, Oracle's current primary conflict is not a lack of AI contracts, but rather the rapid increase in the scale of capital required to acquire and fulfill these contracts. The company has demonstrated its ability to secure large AI infrastructure contracts, but has not yet equally and fully demonstrated that these contracts can generate sufficiently high and sustainable cash flow per common share after deducting data center construction, equipment procurement, long-term leases, power commitments, interest expenses, and potential equity dilution.

This rating downgrade does not directly refute Oracle's AI growth strategy, but it signifies that the company's financial characteristics are shifting. As the importance of OCI and AI infrastructure rises, Oracle can no longer be analyzed solely as a traditional software company with high gross margins and low capital requirements. The company is increasingly exhibiting stronger characteristics of being asset-heavy, long-cycle, customer-concentrated, and financing-sensitive.

I. Why High Growth and Credit Downgrades Can Occur Simultaneously

Oracle's fiscal year 2026 results show that the company's cloud transition is accelerating significantly. Full-year cloud revenue reached $34 billion, up 39% year-over-year, with infrastructure-as-a-service (IaaS) revenue at $18.1 billion, up 77% year-over-year; cloud applications revenue was $15.9 billion, up 11% year-over-year. The proportion of cloud revenue to Oracle's total revenue has increased significantly.

RPO rose from $138 billion at the end of fiscal year 2025 to $638 billion at the end of fiscal year 2026, primarily driven by several large cloud contracts. Oracle also stated that the majority of the RPO increase in the third and fourth quarters of fiscal year 2026 came from large AI contracts, with customer cash advances for GPUs or GPUs provided directly to Oracle totaling approximately $75 billion.

However, the $638 billion is not pure AI orders, nor is it guaranteed cash revenue to be realized in the future. Oracle's accounting definition of RPO includes amounts already invoiced but not yet recognized as revenue, outstanding invoices, and contract amounts to be invoiced and recognized in the future. As of May 31, 2026, the company expects approximately 12% of the RPO to be recognized as revenue in the next 12 months, about 34% in months 13 to 36, another 34% in months 37 to 60, and the remainder over a longer period.

This means that while RPO improves the visibility of future revenue, it cannot on its own answer when cash will be received, whether customers can fulfill their obligations, what the project profit margins will be, and how much capital needs to be invested to deliver these contracts.

The core logic behind S&P's rating downgrade is also not a lack of customer demand for Oracle, but rather that the AI infrastructure business is weakening the company's original business risk and financial risk profiles. S&P stated that it had previously underestimated the scale of investment required for Oracle to expand its AI business and the impact of these investments on its overall credit profile.

S&P also views customer concentration as a significant risk. According to S&P's estimates, OpenAI may account for approximately half of Oracle's current $638 billion RPO; this proportion is an estimate by the rating agency and is not customer breakdown data disclosed by Oracle in its audited financial statements.

Therefore, a stable outlook does not mean risks have disappeared, but rather indicates that a further downgrade is not S&P's current base case. Oracle's management also stated in its 10-K that the company's existing cash, operating cash flows, and financing arrangements are expected to be sufficient to cover working capital, committed capital expenditures, and contractual obligations for at least the next 12 months. It should be noted that this belongs to the company's liquidity judgment, rather than an unconditional guarantee of future liquidity provided by the rating agency.

II. The Capital Intensity of Oracle's Business Model Is Rising

Oracle's traditional database licensing, software support, and enterprise applications businesses feature strong recurring revenue and customer stickiness. Software support contracts are typically invoiced in advance, with revenue recognized over the service period; once a software product is developed, new customers generally do not require Oracle to make concurrent investments in separate, dedicated data centers, power, and computing equipment.

AI infrastructure is different. To provide large-scale GPU computing power, Oracle needs to secure data center space, power, cooling, network, servers, and related infrastructure in advance, and enter into long-term lease and supply agreements. The company's cash investments and contractual liabilities often precede capacity delivery, while revenue is recognized progressively over the contract term or based on actual usage after the capacity is put into service.

This change is already reflected in the cash flows for fiscal year 2026. Oracle's operating cash flow was $31.977 billion, and its reported capital expenditures were $55.663 billion. According to the company's definition—operating cash flow minus capital expenditures—non-GAAP free cash flow was negative $23.686 billion.

It must be emphasized that the $31.977 billion is Oracle's consolidated operating cash flow on a company-wide basis, not the cash flow contributed independently by mature businesses such as database, software support, and enterprise applications. Oracle has not disclosed sufficient data to precisely calculate the independent free cash flow of its core software business.

In addition, operating cash flow for fiscal year 2026 includes approximately $4.6 billion in customer advances that contain a significant financing component. Therefore, the entire operating cash flow cannot be simply viewed as recurring cash naturally generated by mature businesses.

Oracle also disclosed a supplementary metric, namely that the "net cash outflow for capital expenditures" for fiscal year 2026 was $47.726 billion. This metric is derived based on the reported capital expenditures of $55.663 billion, deducting $3.345 billion in short-term financing cash flows related to capital expenditures, and $4.592 billion in customer advances with a significant financing component.

These two capital expenditure metrics serve different purposes. The $55.663 billion represents capital expenditures in the statement of cash flows, used to calculate Oracle's published free cash flow; the $47.726 billion is a supplementary metric used by management to measure the company's actual funding needs. The two cannot be mixed to calculate the same set of free cash flow.

To support its capital expansion, Oracle issued $43 billion in aggregate principal amount of senior notes in fiscal year 2026, with net financing proceeds of approximately $42.7 billion; at the same time, it issued approximately $5 billion of mandatory convertible preferred stock. The company also established an at-the-market (ATM) common stock offering program of up to $20 billion, under which no common shares had been sold as of May 31, 2026.

Mandatory convertible preferred stock is not common stock, but it requires the payment of a 6.5% preferred dividend prior to conversion, and is expected to be mandatorily converted into common stock in January 2029, with the final number of converted shares depending on the stock price at that time and specific terms. Therefore, while this financing instrument can alleviate near-term debt pressure, it will affect the fully diluted common share count in the future.

For fiscal year 2027, Oracle management expects total capital expenditures could reach $90 billion to $95 billion, of which approximately $70 billion will be borne by Oracle itself, and another $20 billion to $25 billion is expected to be reimbursed by customers. Management did not disclose the specific collection timing of the relevant payments.

Oracle also plans to raise approximately $40 billion through debt and equity financing in fiscal year 2027, which includes the previously announced $20 billion common stock ATM program. The company stated that it does not expect to issue additional debt in calendar year 2026, which does not contradict its fiscal year 2027 financing plans, as Oracle's fiscal year extends to May 2027, and the financing plans also include equity instruments.

The aforementioned changes indicate that Oracle has not suddenly lost profitability, but rather that the upfront capital, fixed obligations, and external financing required for incremental revenue have increased significantly. Revenue growth and the deterioration of free cash flow can occur simultaneously during the same period.

III. The $638 Billion RPO Requires Five Layers of Value Transformation

RPO should not be directly multiplied by a valuation multiple to derive equity value. A more reasonable approach is to estimate the probability-weighted after-tax cash flows generated by different contracts over different periods, deduct delivery costs, operating expenses, capital expenditures, leases, working capital, and taxes, and then discount them according to the corresponding risks.

1. Customer Performance and Collection Risks

Large AI contracts can demonstrate that customers have strong computing power demand at the time of signing, but they may also increase customer concentration and counterparty risk.

Oracle explicitly pointed out in its 10-K that if the company overestimates customer demand, or if key customers are unable to pay or fulfill contracts, Oracle could be locked into multi-year data center space, capital expenditure, and financing obligations without obtaining the corresponding revenue. The company also stated that advances, credit reviews, and monitoring procedures cannot completely eliminate customer credit risk, as some customers themselves may have high leverage.

Therefore, what RPO proves is contractual commitment, rather than final utilization, actual collection, or economic return.

2. Timing of Revenue and Cash Conversion

Contracts requiring new AI capacity typically require Oracle to first complete data center, power, network, and equipment deployment before delivery and revenue recognition can begin.

However, not all of the company-wide RPO belongs to new AI infrastructure contracts; it also includes software support, existing cloud capacity, and other services. Therefore, it cannot be assumed that the entire $638 billion RPO must wait for new data centers to be built.

The longer the RPO recognition period, the wider the timing difference between capital investment and cash recovery. Even if the contract is ultimately fulfilled, later cash flows are lower in economic value than near-term cash flows of the same amount.

3. Normalized Margins

GPU infrastructure must bear costs such as power, cooling, network, data center rent, equipment depreciation, maintenance, replacement capital expenditures, and idle capacity, and its profit structure differs from software licensing and maintenance support.

Oracle management stated that the gross margin of the delivered AI capacity in the third quarter of fiscal year 2026 was approximately 32%, higher than the previous guidance of 30%. While this figure indicates that the delivered portion of AI capacity is not necessarily unprofitable, it is merely a metric disclosed by management for specific capacity. It is not the audited gross margin for the entire OCI segment, nor is it a substitute for return on invested capital (ROIC).

Even if the project gross margin reaches 32%, the incremental return on invested capital may still be lower than the cost of capital if obtaining the relevant revenue requires massive upfront investments in equipment, data centers, power, and financing capital. Conversely, if customers bear a significant portion of GPU procurement costs, capacity quickly achieves high utilization, and drives database, network, and storage revenues, the comprehensive return could also be significantly higher than that of a pure GPU leasing business.

4. Allocation of Capital Burden

Oracle disclosed that the hardware portion prepaid or directly provided by customers in large AI contracts totals approximately $75 billion. This reduces Oracle's need to raise the entirety of GPU procurement funds on its own and serves as an important risk buffer.

However, the $75 billion is neither fiscal year 2026 revenue nor equivalent to Oracle having already received $75 billion in cash, as it includes equipment purchased directly by customers and provided to Oracle.

This figure also cannot be directly added to the $20 billion to $25 billion expected to be reimbursed by customers in fiscal year 2027. The former represents customer hardware support accumulated in large AI contracts, while the latter is management's forecast of the customer-reimbursed portion of total capital expenditures for fiscal year 2027. The two differ in both timeframe and contract scope.

5. Value Retention per Share

Smooth contract execution does not mean corporate revenue can be proportionally converted into value per common share.

Revenue must also bear interest, leases, equipment updates, preferred dividends, preferred stock conversions, common stock issuances, and the claims of other capital providers. Therefore, for common shareholders, what truly needs to be measured is how much fully diluted free cash flow per share can be generated after the contracts go through delivery, collection, profit, capital investment, and financing structures.

IV. Quadruple Maturity Mismatch Is the Core Risk of the Balance Sheet

As of May 31, 2026, notes and other borrowings in Oracle's balance sheet totaled $129.541 billion, with the current portion at $7.199 billion and the non-current portion at $122.342 billion.

During the same period, Oracle's recognized operating lease liabilities were approximately $30.190 billion, and finance lease liabilities were approximately $7.701 billion. Borrowings, operating leases, and finance leases belong to different accounting line items and cannot be directly combined into a single "debt" figure without adjustments, but they all impact the company's fixed payment capacity and financial flexibility.

In addition, Oracle has $260 billion in additional uncommenced lease commitments that are not yet recorded on the balance sheet, the vast majority of which relate to data center arrangements. These leases are typically expected to commence from the first quarter of fiscal year 2027 through fiscal year 2029, with terms spanning 15 to 19 years.

The $260 billion represents undiscounted nominal contractual commitments, rather than currently recognized new debt, and cannot be directly added to the $129.541 billion in borrowings. However, it represents large-scale, long-term fixed obligations that will gradually take effect in the future.

1. Mismatch between customer contracts and lease terms

OpenAI's official disclosure shows that its partnership with Oracle, reached in July 2025, involves up to 4.5 gigawatts of new Stargate capacity, with the scale of cooperation between the two parties exceeding $300 billion over the next five years.

Lease terms for some of Oracle's new data centers are as long as 15 to 19 years. The company explicitly acknowledged in its 10-K that the terms, renewal options, and price adjustments of long-term data center leases typically do not fully match customer contract terms and pricing; if a customer does not renew, Oracle may be unable to sublease, reconfigure, or transfer the associated capacity on acceptable terms.

However, public information does not prove that all of the $260 billion in lease commitments are built for OpenAI, nor does it disclose the one-to-one mapping between each customer contract and each data center lease. Therefore, the five-year partnership and the 15-to-19-year leases can only serve to illustrate potential maturity mismatches on the overall balance sheet, rather than being presented as a verified one-to-one contractual relationship.

2. Mismatch between economic life of equipment and infrastructure terms

Oracle's accounting useful lives for computers, networks, machinery, and equipment typically range from 1 to 6 years, with estimated useful lives for major servers and network equipment at approximately 6 years; buildings and improvements can have useful lives of up to 40 years.

Accounting useful lives do not equate to the actual economic life of specific GPUs, but they reflect that the duration of data centers, buildings, power, and long-term leases may be significantly longer than the time the equipment remains highly competitive.

Older GPUs do not need to fail completely to create economic pressure. If next-generation chips offer higher performance, lower energy consumption, and lower cost per unit of compute, the realized pricing, utilization, and profitability of older equipment may decline.

The counterargument is that older equipment can still be repurposed for inference, enterprise workloads, or other lower-performance needs, and thus does not necessarily lose all residual value. Since Oracle does not disclose the utilization, contract pricing, and secondary market value of different GPU models, this risk cannot be precisely quantified at present.

3. Mismatch between cash investment and revenue recognition

Oracle must pay for data centers, equipment, power, and financing costs before the associated capacity generates revenue. The company explicitly warns that if it overestimates demand or if customers default, it could bear multi-year data center, capital expenditure, and financing obligations without corresponding revenue.

This means that the phase of strongest growth in contracts and RPO may also be the phase of greatest pressure on free cash flow and external financing. Order growth in itself does not prove that balance sheet risk is declining.

4. Mismatch between fixed obligations and long-term demand uncertainty

As of May 31, 2026, Oracle also had $13.309 billion in unconditional purchase and other obligations, primarily related to data center power arrangements. After the end of the fiscal year, the company added another $19 billion in cloud infrastructure asset purchase commitments, starting in fiscal year 2027 with a term of five years.

These commitments further indicate that Oracle's AI expansion depends not only on capital expenditures and leases but also involves long-term power and infrastructure procurement obligations.

Large customer contracts may include prepayments, minimum purchase commitments, or committed capacity, so a portion of revenue during the contract term is not entirely uncertain. However, uncertainties remain regarding whether customers will renew, whether capacity can be subleased after contracts expire, whether long-term utilization can be maintained, and how future compute pricing will evolve.

V. AI Infrastructure Must Pass the Return on Invested Capital (ROIC) Test

AI infrastructure revenue can be approximately understood as the combined result of deliverable capacity, actual or contracted utilization, and realized price per unit of compute.

Actual contracts may include terms such as reserved capacity, minimum spend, fixed fees, different GPU models, networking and storage services, and customer-provided equipment. Therefore, this relationship can only serve as an analytical framework rather than an accounting formula.

Economic returns must also account for power, cooling, networking, rent, equipment refreshes, working capital, idle capacity, and cost of capital. OCI revenue growth itself is not the ultimate standard for value judgment; whether the incremental return on invested capital (ROIC) in the mature phase is higher than Oracle's cost of capital is far more critical.

Currently, public data is insufficient to directly calculate OCI's standalone normalized return on invested capital. Oracle does not regularly disclose OCI's standalone invested capital, full operating income, data center utilization, contract-level capital expenditures, or the specific allocation of customer-provided assets. Consequently, external investors can only make indirect estimates using cloud revenue, management-disclosed gross margins for AI capacity, consolidated profit margins, depreciation, fixed assets, capital expenditures, and operating cash flow.

AI efficiency gains are not a one-way positive or negative for Oracle. Stanford University's 'Artificial Intelligence Index Report 2025' shows that the cost of inference to reach the GPT-3.5 level fell by more than 280-fold between November 2022 and October 2024, with AI hardware costs declining by an average of about 30% annually, and energy efficiency improving by an average of about 40% annually.

This means that the compute power and costs required to complete the same workload may continue to decline, thereby putting downward pressure on the price per unit of compute.

On the other hand, falling costs may expand the applicability of AI and stimulate more usage. Total compute revenue ultimately depends on the co-movement of the number of workloads, the compute power required per workload, and the price per unit of compute.

If the growth in the number of workloads outpaces efficiency gains and price declines, Oracle may still benefit from overall demand growth; if efficiency improvements outrun application expansion, some new capacity may face utilization and pricing pressures.

The economic characteristics of training and inference may also differ, but Oracle does not disclose the specific ratio of training versus inference within OCI revenue, nor does it disclose the respective margins and utilization rates of these two types of workloads. Therefore, industry-level inferences cannot be directly treated as Oracle's verified business structure.

VI. Multicloud Database Is Oracle's Higher-Quality Bull Case

If OCI ultimately functions primarily as a GPU compute rental business, its valuation logic will be closer to that of capital-intensive infrastructure companies, making it more sensitive to utilization, equipment lifespan, financing costs, and price competition.

A more strategically valuable scenario for Oracle is to leverage AI infrastructure and multicloud services to strengthen its database and enterprise application ecosystem.

Oracle's multicloud architecture allows customers to use Oracle Database services within environments like OCI, Microsoft Azure, AWS, and Google Cloud. Consequently, customers do not need to migrate all of their workloads to OCI first, and Oracle can still generate revenue through database consumption, data management, and enterprise applications.

The potential value of this strategy lies in Oracle's ability to continue dominating the enterprise core database and data tier, regardless of which public cloud customers primarily use. If database usage further drives middleware, analytics, ERP, healthcare applications, and AI agents, the comprehensive return on infrastructure investment could be higher than what would be concluded from looking at the GPU leasing business in isolation.

Oracle stated that its Multicloud AI Database business grew by 404% in the fourth quarter of fiscal year 2026. While this figure demonstrates strong business momentum, the company does not separately disclose its revenue base, absolute revenue, gross margin, operating profit, and invested capital. Therefore, one cannot conclude that it already constitutes a material source of earnings based solely on the 404% growth rate.

The counterargument is that multicloud database growth does not necessarily translate proportionally into OCI compute revenue. Customers using Oracle Database in AWS, Azure, or Google Cloud may strengthen database revenue and customer stickiness, but they will not necessarily migrate all of their infrastructure consumption to OCI.

Therefore, Oracle's bull case requires the simultaneous verification of two conditions: whether OCI itself can generate a reasonable return on capital, and whether AI and multicloud businesses can genuinely drive higher-quality database and application profits.

VII. Credit Ratings Are a Financial Constraint on Oracle's Pace of Expansion

Potential downgrade triggers listed by S&P include Oracle's adjusted leverage remaining sustained above 4.5x, or if the rating agency no longer believes the company has a credible path to returning to positive free operating cash flow by fiscal year 2029.

The 4.5x and fiscal year 2029 mentioned here represent analytical conditions for rating purposes, rather than legal default triggers for bonds or covenant thresholds for bank loans.

The free cash flow reported by Oracle is operating cash flow minus reported capital expenditures; the free operating cash flow used by S&P adjusts for leases, capitalized interest, working capital, and other items based on its rating methodology. Although their names are similar, they cannot be directly interchanged.

If Oracle were to be downgraded another notch by S&P to BB+, it would enter speculative-grade territory in S&P's rating system. Generally, this could widen the credit spreads demanded by investors, reduce flexibility for new debt issuance and refinancing, and prompt certain investment-grade-constrained investors to reduce their holdings.

However, the actual volume of passive selling depends on other rating agencies and specific fund and index rules; it cannot be assumed that a downgrade by S&P alone would lead to all investment-grade bond funds simultaneously selling Oracle bonds.

Negative reflexivity could manifest as: a rating downgrade leading to higher financing costs; a higher cost of capital reducing project net present values and prolonging cash flow pressures; and the company subsequently raising equity or other financing, resulting in further dilution for common shareholders.

However, equity financing also has positive credit implications. While issuing common stock or converting preferred stock dilutes existing shareholders, it can reduce incremental debt and lower leverage.

Positive reflexivity, on the other hand, requires an increase in customer prepayments and equipment support, on-time delivery of data centers, smooth conversion of RPO, and growth in operating profit and operating cash flow that outpaces debt.

Overall, the stable outlook indicates that another near-term downgrade is not S&P's current base case. However, if capital expenditures continue to significantly exceed guidance, customer payments are delayed, or the path to returning to positive cash flow by fiscal year 2029 loses credibility, the risk of another downgrade will rise significantly.

VIII. Valuation Needs to Shift from P/E Ratio to Balance Sheet Look-Through Analysis

The price-to-earnings (P/E) ratio under standard accounting measures already incorporates the impact of recognized depreciation, interest, lease expenses, and preferred stock dividends on common earnings, as well as the diluted share count for the period.

Therefore, the issue is not that the P/E ratio completely ignores these costs, but rather that it cannot adequately reflect future capital expenditures, leases that have not yet commenced, equipment replacements, cash flow timing mismatches, unexecuted common stock issuances, and the future actual converted shares of mandatory convertible preferred stock.

To value Oracle, equity value can be split into three parts:

Oracle Common Equity Value = Value of Mature Software Business + Risk-Adjusted OCI and AI Growth Options - Balance Sheet and Per-Share Dilution Costs

Mature database, software support, and enterprise application businesses can provide an enterprise value anchor, but they do not represent an absolute stock price floor immune to AI expansion. Cash generated by mature businesses may still be used for data center construction, and Oracle does not disclose standalone free cash flow and standalone capital expenditures for its mature software business.

Similarly, after deducting the valuation of mature businesses from enterprise value, the remaining value cannot be entirely and directly attributed to OCI; it may also include Oracle Health, hardware, services, enterprise applications, multi-cloud synergy, corporate headquarters costs, tax differences, and segment valuation errors.

To reverse-value RPO, the following questions must be answered: whether contracts can be fulfilled and collected, whether capacity can be delivered on time, whether revenues can generate reasonable operating margins, whether profits can translate into cash after deducting capital investments, and whether this cash can be converted into value per common share after financing and dilution.

Due to the lack of customer-level margins, contract cancellation clauses, contract-level capital expenditures, data center utilization rates, and collection schedules, external investors cannot precisely calculate a public 'RPO-to-equity conversion rate'.

Valuation methodologies must also avoid double counting. If unlevered free cash flow and the overall cost of capital are used, interest should not be deducted from cash flows and then the same financing cost deducted again at the enterprise value stage; depreciation and replacement capital expenditures also cannot both be fully deducted twice as the same economic cost.

The $260 billion in lease commitments that have not yet commenced also cannot be fully deducted from market capitalization all at once. This figure represents undiscounted nominal contractual payments and corresponds to future data center right-of-use assets and potential revenues. If a model has already incorporated lease expenses into future profits and cash flows, an additional full deduction of lease commitments would result in double counting.

One may choose to discount lease payments and treat them as debt-like, or continuously incorporate lease expenses into future cash flows, but the valuation methodology must remain consistent.

Before the path to positive free cash flow is validated, Oracle is better viewed as a high-growth, capital-intensive AI infrastructure investment that depends on customer credit and financing conditions, rather than a traditional low-risk, defensive software asset.

IX. Three Conditional Investment Scenarios

Bull Case Scenario

The bull case scenario requires RPO to convert as planned, customer reimbursements, prepayments, and customer-provided equipment to materialize on time, and new data centers to achieve high utilization rates relatively quickly.

Meanwhile, OCI needs to drive database, multi-cloud, and enterprise application businesses, making the consolidated quality of profits higher than that of a pure-play GPU infrastructure business. Operating cash flow growth must outpace debt and capital expenditures, the path to restoring positive free operating cash flow before fiscal year 2029 must remain credible, and external financing and common stock dilution must be lower than market expectations.

Under this scenario, the rating downgrade may be viewed as a transitional cost of Oracle achieving AI infrastructure status, and the company's valuation could regain a premium for its software and platform attributes.

However, Oracle has not currently guided that capital expenditures will peak in fiscal year 2027 or 2028. Instead, the company expects the upward trend in capex to continue into fiscal year 2027 and several subsequent fiscal years. Therefore, an earlier inflection point for capital expenditures can only serve as a bull-case assumption, rather than a management commitment.

Base Case Scenario

In the base case scenario, OCI and total revenue continue to grow relatively quickly, but margins and free cash flow improve more slowly than revenue growth.

There are timing delays in customer reimbursements, the company continues to rely on debt, preferred stock, common stock, and other financing instruments to support construction, and the BBB- rating is maintained.

Under this scenario, Oracle does not necessarily experience a fundamental failure, but enterprise revenue may grow faster than fully diluted cash flow per share, with valuation constrained over the long term by capital intensity, financing costs, and dilution risk.

Bear Case Scenario

The bear case scenario includes key customers delaying, scaling back, or renegotiating contracts, data center commissioning occurring more slowly than the onset of lease obligations, utilization rates and realized pricing for computing power falling below expectations, and capital expenditures exceeding the $90 billion to $95 billion guidance.

If customer reimbursements are delayed, adjusted leverage remains persistently above 4.5x, and the path to restoring positive free operating cash flow before fiscal year 2029 disappears, Oracle could face the risk of a further downgrade to BB+.

At the equity level, the stock may simultaneously suffer from downward earnings revisions, equity dilution, and valuation multiple contraction.

Even under a bear case scenario, Oracle's options are not limited to slowing down construction or issuing additional common stock. The company could also require customers to increase prepayments or provide equipment, utilize project financing, lease financing, joint venture arrangements, asset partnerships, or sell off certain assets, and cut other capital allocations. However, should the credit environment deteriorate, the costs and negotiating terms of these alternatives could also worsen.

X. Metrics Investors Should Monitor

First, investors should distinguish between total capital expenditures, customer reimbursements, customer prepayments, related short-term financing, and Oracle's own actual cash outlays, rather than focusing solely on a single capital expenditure figure.

Second, there is a need to assess the quality of RPO conversion, including the expected recognition ratio over the next 12 months, actual growth in cloud revenue, changes in deferred revenue, contract delays, and whether customer funding support materializes on time.

Third, the path to restoring positive free operating cash flow before fiscal year 2029 needs to be continuously tested. Whether operating cash flow can gradually catch up with capital expenditures and financing needs is an important condition for maintaining the current investment-grade rating.

Fourth, because OCI's standalone return on invested capital is not public, investors can only use OCI revenue growth, AI capacity gross margins, consolidated margins, depreciation, fixed assets, capital expenditures, and operating cash flows as proxy metrics.

Fifth, it is necessary to track new borrowings, recognized leases, uncommenced lease commitments, $19 billion in new purchase commitments, preferred stock conversions, and actual issuances of ATM common stock.

Finally, accounting and valuation methodologies must remain consistent. The reported capital expenditures of $55.663 billion carry a different meaning from the net cash outflow for capital expenditures of $47.726 billion; similarly, Oracle's self-defined free cash flow and S&P-adjusted free operating cash flow are not directly interchangeable.

Conclusion

Oracle's current fundamentals cannot be simply summarized as certain growth following validated AI demand, nor can they be simply categorized as a strategic failure resulting from credit deterioration.

The confirmed facts are that Oracle has secured a large volume of long-term contracts, OCI infrastructure revenue is growing rapidly, and customers are bearing a portion of the capital through prepayments and direct equipment provisioning; at the same time, the company's capital expenditures, borrowings, long-term leases, power obligations, and external financing requirements have increased significantly, free cash flow has turned deeply negative, and its credit rating has dropped to the lowest tier of investment grade.

This means that the investment thesis for Oracle has shifted. In the past, the market focused primarily on database moats, software renewals, and operating margins; in the future, it must also simultaneously assess data center utilization, customer credit, contract durations, equipment upgrades, cost of capital, and fully diluted value per share.

Oracle has proven its ability to win large-scale AI contracts. What needs to be validated in the next phase is whether these contracts, after deducting construction, leasing, power, equipment replacement, financing, and dilution, can generate cash flow per share that is above the cost of capital, sustainable, and attributable to common shareholders.

This is not only an investment question for Oracle itself but also represents the core test that the entire AI infrastructure industry must face: Real and massive AI demand does not automatically translate into value for common stock; contract sizes only truly carry equity significance when revenue growth ultimately translates into cash returns that exceed the cost of capital.