Oracle (ORCL): Can AI Contracts Become Per-Share Cash?
Oracle's FY2027 Q1 revenue was US$19.345 billion, up about 30%, and IaaS revenue grew 121%. But capex reached US$28.499 billion, company-defined free cash flow was negative US$5.396 billion, and RPO stood at US$664 billion. We test whether the contracts can turn into per-share cash through GPU depreciation life, customer prepayments, US$288 billion of not-yet-commenced leases and equity dilution: cloud growth has real business value, but capital recovery is not yet sufficiently proven.
排除
📌 Key Conclusions
- FY2027 Q1 revenue was US$19.345 billion, up approximately 30% year over year, and IaaS grew 121%; capital expenditures were US$28.499 billion, however, and company-defined free cash flow was negative US$5.396 billion.
- FY2026 operating cash flow was US$31.977 billion and capital expenditures were US$55.663 billion; simplified free cash flow moved from positive US$11.807 billion in FY2024 to negative US$23.686 billion.
- RPO was US$664 billion, with approximately 13% (approximately US$86.3 billion) expected to be recognized over the next 12 months; US$11.363 billion of Q1 operating cash flow of US$23.103 billion came from customer prepayments, approximately 49%.
- The economic value of GPUs, six years of straight-line depreciation, customer contracts of one to five years, and data center leases of 15 to 19 years may not move in sync; US$288 billion in leases not yet commenced is not recognized debt, but is an important future constraint.
- Reports that Microsoft is winding down some older GPU rentals coexist with Oracle's response and cannot yet be taken as evidence of default or impairment; this research considers cloud growth to have real commercial value, but capital recovery has not been sufficiently demonstrated.
Oracle is expanding from a company centered on databases, enterprise applications, and support services into a cloud provider requiring large-scale data center and GPU construction. As of November 9, 2026, the latest complete quarterly report was FY2027 Q1: revenue was US$19.345 billion, up approximately 30% year over year, and cloud infrastructure (IaaS) revenue rose 121%; but capital expenditures reached US$28.499 billion and company-defined free cash flow was negative US$5.396 billion. The question has therefore shifted from whether Oracle can win AI orders to whether those orders can leave enough cash per share after equipment replacement, long-term leases, interest, and equity dilution.Company FY2027 Q1 results
My view is: Oracle's enterprise customer base and cloud growth have real commercial value, and its AI infrastructure transformation is already reflected in revenue, but capital recovery has not been sufficiently demonstrated. Core operating profits and customer prepayments provide important support, but borrowing, long-term leases, and share issuance expose common shareholders to greater funding and performance risks. This is not a finding of an immediate debt repayment crisis, nor a claim that RPO orders do not exist; the point is that large contracts or a single quarter's revenue cannot replace a complete assessment of per-share investment returns.
⚠️ Research warning: customer nonrenewal, equipment economic life, and off-balance-sheet commitments may act together. On November 6, 2026, media reported that Microsoft was winding down an existing arrangement to rent some AI capacity from Oracle involving an older generation of GPUs. This warning requires verification in subsequent financial reports: it cannot be directly inferred to mean default, wholesale cancellation, or an impairment already incurred, but it offers a concrete example of the gap between how long a customer rents and how long equipment is depreciated for accounting purposes; Oracle responded that demand, including for older-generation GPUs, remained strong (see Section Four for details).Bloomberg report
Core conclusion: Oracle's AI investment must pass four timing tests at once. The rate at which GPU economic value declines, equipment's accounting depreciation period, customer contract and collection periods, and financing repayment and data center lease terms may not move in sync. If any one component lags cash outflows, even continued revenue growth may be accompanied by greater borrowing, equity dilution, or impairment risk. What must really be verified is whether each investment can recover cash after operating, financing, and exit costs before the equipment loses competitiveness, rather than merely how large the contracts are.
1. The industry first: AI cloud demand is real, but economic returns still depend on supply, demand, and contracts
Enterprise AI infrastructure is a long chain. Upstream suppliers of chips, memory, networking, and power equipment provide hardware; midstream cloud providers handle procurement, construction, power, networking, computing, and operations; downstream are model developers, enterprise software companies, and businesses using AI. Oracle provides OCI computing capacity in the middle and connects with end enterprises through databases and enterprise applications.
Demand starts with large-model training, but also includes inference, enterprise data analysis, and migration of existing workloads. Contracts with large model developers can sharply increase near-term orders, but their ultimate funding still comes from customer revenue, external financing, or other cash sources; a cloud provider winning a contract and its downstream customer being able to pay over the long term are therefore separate questions to verify. As AI models become more efficient, computing needed per use may fall: if lower prices create new uses, total demand may still rise; otherwise, utilization of existing capacity may weaken.
Competition extends beyond cloud capacity. AWS, Microsoft Azure, Google Cloud, and other providers compete on enterprise services, GPU supply, cost, and delivery times; large customers may also build their own infrastructure or adopt multicloud strategies. Oracle can cross-sell through its databases, applications, and OCI, but database customers will not necessarily give Oracle all their new workloads. This research views that customer base as an advantage in winning opportunities, not evidence of monopoly or unconditional pricing power.
2. How Oracle makes money: renewals in established businesses support expansion in new ones
Oracle databases have long been embedded in enterprise transaction systems, financial processes, and applications. When customers change systems, they must not only migrate data but also retest applications, permissions, reliability, and staff processes. These costs give existing customers reasons to renew and continue using the products, while generating revenue for support services and cloud databases. The moat comes from switching costs created jointly by data, software, and workflows, but alternatives, multicloud deployments, and customer bargaining power still limit pricing.
Cloud applications sell software such as ERP by subscription; IaaS provides customers with computing, storage, and other infrastructure services. Although both are classified as cloud, their economics differ: software subscriptions can serve more customers on existing products, while increased cloud computing often requires additional equipment, space, power, and networking. As fast-growing IaaS becomes a larger share, the group may see revenue growth and rising capital intensity at the same time, and past cash conversion assumptions for pure software cannot be applied directly.
In the Q1 disclosures, for example, IaaS revenue was approximately US$7.4 billion and SaaS approximately US$4.2 billion, together approximately US$11.6 billion; traditional software revenue was US$5.55 billion, down 3% year over year. The established businesses remain important sources of cash and customers, but the company did not fully break down how much incremental cloud revenue came from existing database customers and how much from large AI contracts in this data. There is insufficient evidence to interpret all IaaS growth as a natural extension of the old moat.FY2027 Q1 results
3. Latest results: growth and profitability improved, but first establish the measurement basis
Three complete fiscal years help distinguish what Oracle's cloud transformation has changed. Oracle's fiscal year ends May 31; FY2024 through FY2026 below are audited full years, in US$100 million except EPS. FY2027 Q1 is a subsequent single quarter, shown separately and not compared directly with full years.
Metric | FY2024 | FY2025 | FY2026 | Comparison |
|---|---|---|---|---|
Revenue | 529.61 | 573.99 | 673.57 | Up 8.4% in 2025 and 17.3% in 2026; growth accelerated |
Total cloud revenue | 197.74 | 245.06 | 339.89 | Increased by US$14.215 billion over two years; share of revenue rose from 37.3% to 50.5% |
Of which: cloud infrastructure IaaS | 68.40 | 102.34 | 181.01 | Up 49.6% and 76.9% respectively; became the strongest growth engine |
Of which: cloud applications SaaS | 129.34 | 142.72 | 158.88 | Up 10.3% and 11.3% respectively; steadier growth |
Operating income (GAAP) | 153.53 | 176.78 | 206.06 | Up 15.1% and 16.6% respectively, but returns on individual GPU projects were not disclosed |
Operating margin (calculated) | 29.0% | 30.8% | 30.6% | Down approximately 0.2 percentage points in 2026 from the prior year |
Net income (GAAP) | 104.67 | 124.43 | 170.87 | Up 18.9% and 37.3% respectively; 2026 includes gains on investment disposal |
Diluted EPS (US dollars, common stock) | 3.71 | 4.34 | 5.83 | Up 17.0% and 34.3% respectively; share count and nonoperating gains warrant attention |
Cash flow from operations | 186.73 | 208.21 | 319.77 | Up 53.6% in 2026; includes a US$4.592 billion change in prepayments with a significant financing component |
Capital expenditures (cash flow statement) | 68.66 | 212.15 | 556.63 | Increased US$48.797 billion over two years, far more than the increase in operating cash flow over the same period |
Simplified free cash flow: operating cash flow less capital expenditures | 118.07 | negative 3.94 | negative 236.86 | Turned negative from positive; expansion requires other funding sources |
Interest expense | 35.14 | 35.78 | 45.99 | Up 28.5% in 2026, reflecting increased borrowing |
Source: FY2026 Form 10-K consolidated income statements, cash flow statements, and cloud revenue notes. Free cash flow is calculated using the same definition as the company and is a non-GAAP measure; FY2024 figures are recalculated from the financial statements. EPS is diluted earnings per common share; net income is the amount shown in the table and should not simply be divided by the period-end share count.
The revenue mix is changing faster than the traditional software business. Across the three years IaaS rose from US$6.840 billion to US$18.101 billion, approximately 165%; SaaS rose from US$12.934 billion to US$15.888 billion, approximately 23%. IaaS was initially smaller than SaaS but had surpassed it by FY2026; cloud's share of total revenue also rose from approximately 37% to 51%. This supports the view that cloud has become the company's growth engine, but the higher infrastructure share also means new growth increasingly depends on equipment and data centers. This is an observation about revenue mix and does not imply identical investment returns for every product.
Operating income and net income are both growing, but not all net income growth can be interpreted as improvement in core operations. FY2026 operating income rose approximately 16.6% year over year, while operating margin slipped from 30.8% to 30.6%; revenue growth and margin expansion do not necessarily occur together. FY2026 net income was also affected by nonoperating items including approximately US$2.7 billion in gains on the sale of the Ampere investment, so not all net income growth can be attributed to improvement in day-to-day cloud operations.10-K discussion of nonoperating income and expenses
The change in free cash flow marks the turning point in capital allocation. From FY2024 to FY2026, operating cash flow increased approximately US$13.304 billion, while capital expenditures increased approximately US$48.797 billion, turning free cash flow from positive US$11.807 billion to negative US$23.686 billion. FY2026 operating cash flow included a US$4.592 billion change in customer prepayments with a significant financing component; it is actual cash received but carries future performance obligations and cannot be assumed to recur annually. A widening cash gap does not mean the investment has failed; it means expansion requires other funding sources, and whether new assets will deliver sufficient future returns remains to be verified.
Latest quarter: comparison with the prior-year period
Metric | FY2026 Q1 | FY2027 Q1 | How to interpret it |
|---|---|---|---|
Revenue | US$14.926 billion | US$19.345 billion | Up approximately 30%, driven by cloud |
GAAP operating income | US$4.277 billion | US$6.728 billion | Up approximately 57%; timing of costs and depreciation still warrants attention |
GAAP diluted EPS | 1.01 | 1.56 | For common shareholders; affected by share count and preferred stock, among other factors |
Operating cash flow | US$8.140 billion | US$23.103 billion | The quarter includes prepayments with a significant financing component |
Cash capital expenditures | US$8.502 billion | US$28.499 billion | Expansion of data center and equipment investment |
Company-defined free cash flow | negative US$362 million | negative US$5.396 billion | Operating cash flow less cash capital expenditures |
Sources: Oracle FY2027 Q1 release, FY2027 Q1 10-Q. Q1 revenue, costs, and earnings are under GAAP; free cash flow is a company non-GAAP measure, not GAAP operating cash flow.
The company's operating margin this quarter was approximately 34.8%, up from approximately 28.7% a year earlier. This is evidence of an actual improvement in profitability, but does not mean every new data center will earn the same return. Construction, commencement of service, and full utilization may span several quarters, and depreciation, electricity, and financing costs are recognized at different times; observation must therefore continue, and this quarter's operating margin cannot be used to value all new assets.
Depreciation expense also rose from US$1.351 billion a year earlier to US$3.156 billion, reflecting the cost burden of a larger asset base. If revenue and utilization of new capacity grow faster, depreciation may be absorbed; if delivery lags or customer usage is insufficient, depreciation and fixed expenses may compress profits. These are inferences under two sets of conditions, not predictions of next quarter's results.10-Q cash flow and income statements
Six years of depreciation: can GPUs really earn for six years?
⚠️ Method clarification: Oracle uses the straight-line method, not accelerated depreciation with higher early charges. The FY2026 10-K accounting policy states that fixed assets are depreciated on a straight-line basis over their estimated useful lives only once they are in the intended condition and location for use; paying for equipment still under construction does not mean depreciation has begun, and payment alone does not start depreciation on all construction in progress. The note lists estimated useful lives of one to six years for “computers, networks, machinery and equipment,” principally servers and network equipment estimated to last six years. The company does not individually disclose actual depreciation periods, utilization, or retirement dates for all GPUs in this note, so one cannot directly assert that every GPU is depreciated over six years.
Why does the straight-line method require special attention? If equipment loses commercial value faster in its early years because of technological progress and customers moving to newer chips, the straight-line method still allocates cost relatively evenly across years; compared with an accelerated method with higher early expenses, early reported profits and asset carrying values may be higher. But if older equipment has stable demand and relatively even use value, the straight-line method may accurately reflect its use pattern. Use of the straight-line method alone cannot establish that the statements are wrong, and the replacement cycle cannot be equated with the disposal date.
For example (not company data), assume equipment costs US$10 billion, has zero residual value, lasts six years, and is placed in service at the start of the year. The straight-line method recognizes approximately US$1.667 billion in each of the first two years, totaling US$3.333 billion. Using the double-declining-balance method solely for comparison would produce approximately US$3.333 billion and US$2.222 billion in the first two years, totaling US$5.556 billion, approximately US$2.222 billion more than the straight-line method. This is an illustrative calculation, not Oracle accounting data or a determination of the appropriate method; if an accelerated method is used, remaining carrying value and residual value must still be handled according to the rules. The two methods change only the timing of expenses for the same equipment cost, not the cash outlay at purchase.
Oracle has also extended estimated lives in recent years: in FY2023 it extended servers from four to five years, and in FY2025 extended servers and network equipment from five to six years. The company disclosed that this FY2025 change in estimate reduced operating expenses by approximately US$733 million, increased net income by approximately US$573 million, and increased diluted EPS by approximately US$0.20. This is the effect of the company's change in estimate for equipment then in service, not an estimate of its effect on all new FY2027 GPUs. Extending estimated lives did increase reported earnings, but for accounting classification it is a change in accounting estimate of useful life, not properly described directly as a change in accounting policy, and the extension alone cannot establish a violation of standards.Oracle FY2025 10-K
Six years is not the only industry practice. Microsoft listed two to six years for servers and network equipment in FY2026; Amazon listed five to six years at the end of 2025 and shortened some equipment from six years back to five in 2025. Therefore, this does not establish that six years is inconsistent with current practice among major cloud providers, but it does not mean six years is reasonable for every batch of Oracle AI GPUs. Amazon's shortening for some equipment highlights the need to assess individual assets and technology changes rather than assume all GPUs have the same life.Microsoft annual report; Amazon 2025 10-K
Accounting life and actual service should be considered on three levels. Physical life concerns whether equipment can keep operating and be maintained; technology-relevant life concerns whether hardware can support customers' workloads; economic life concerns whether use remains worthwhile after electricity, maintenance, and necessary costs are deducted from rental revenue. A new generation of GPUs does not automatically make old machines scrap; older GPUs may continue earning if they can handle inference or other work. Conversely, even if hardware still operates, its economic life may be shorter than the accounting estimate if power consumption is too high, performance does not meet customer needs, or prices fall. Public information is insufficient to determine the average service life or actual rental prices of Oracle GPUs.
Consider another simplified example that is not a company forecast: assume depreciable equipment costs US$10 billion, has zero residual value, and is put into use immediately. Straight-line depreciation over six years produces approximately US$1.667 billion in annual expense; over four years it is US$2.5 billion annually, a difference of approximately US$833 million per year for the first four years. All else equal, a shorter life reduces reported operating income in those four years; but this is only a difference in the timing of accounting allocation, and the depreciation period alone does not change the US$10 billion cash originally paid. Actual taxes, timing of use, residual values, and asset mix may differ, so the example cannot be applied as Oracle's actual EPS impact.
More important is the replacement cycle. If equipment must be replaced in its fourth year to serve customers, it will still have undepreciated carrying value, and the company must assess remaining useful life, disposal, and impairment in light of actual circumstances; conversely, if old equipment still has sufficient customers and positive cash contribution, a six-year estimate may be reasonable. The sequence is to examine utilization, prices, electricity costs, maintenance, secondhand values, and subsequent replacement spending first, then whether the company's depreciation estimates or impairment reflect those changes. Oracle's disclosures are not yet sufficient to complete this step-by-step verification.
Accordingly, this research lists the six-year estimate combined with the straight-line method as an earnings-quality warning to monitor. Recent media reports that Microsoft is no longer renewing some older Oracle AI capacity (see Section Four) make it more important to watch whether equipment can be transferred to other customers and whether rents cover electricity and maintenance costs. Available information does not yet prove that the equipment must be replaced early, that the six-year estimate is unreasonable, or that Oracle has incurred an impairment.Bloomberg report
Core risk: who bears the difference when four clocks do not align?
Oracle's 2026 annual report states that Oracle Cloud application and infrastructure contracts generally run one to five years, with renewal at the customer's option; the company's principal servers and network equipment have estimated useful lives of six years. As of the end of August 2026, most data center lease commitments not yet commenced ran fifteen to nineteen years. These three sets of data cover different contracts and assets and are not matched data for the same batch of GPUs, the same customer, and the same data center; individual large AI contracts may also differ from the general terms. They indicate a structural risk worth monitoring, but are insufficient to quantify the actual mismatch.Oracle FY2026 annual report; FY2027 Q1 10-Q
The first clock is economic value. A GPU that still operates cannot necessarily be rented at its original price. If new-generation equipment provides services at a lower cost per unit of computing, older equipment rents or utilization may be pressured; but demand, power constraints, and existing workloads may also allow older equipment to remain profitable. Public market prices, short-term promotions, and particular long-term contracts have different equipment, supply, and service terms; a single promotional discount cannot be used to calculate Oracle's realized rents, nor can all GPU rents be assumed to fall at a fixed rate every year.
The second clock is accounting depreciation. The six-year straight-line method allocates depreciable cost roughly evenly, but does not guarantee that equipment's economic value declines evenly. If early rents and utilization decline faster, the gap between carrying value and future cash recovery may widen; impairment still requires testing under the applicable asset grouping and accounting rules. Extending depreciation increases current accounting earnings but does not reduce cash spent on equipment purchases.
The third clock is customer contracts and collections. Contract value, enforceable minimum commitments, prepayments, performance periods, and customers' actual ability to pay are distinct concepts. If a contract expires without renewal, Oracle must transfer the equipment to other customers at a suitable price and time; if customer payment or delivery conditions become problematic, contractual rights may not immediately become cash. Reports that Microsoft is winding down some older capacity arrangements (see Section Four) are a warning to monitor re-rental capability, but cannot be treated as proof of default or impairment.
The fourth clock is financing and leases. Debt maturities, interest, and data center rent do not automatically disappear when GPU rents fall. A data center can serve different generations of equipment over a longer period, so a fifteen-to-nineteen-year lease cannot be directly treated as a loss against six-year depreciation for one batch of GPUs; but if subleasing or redeployment goes poorly, fixed commitments must still be borne by Oracle's other revenue or new funding.
Consider a hypothetical example: a batch of equipment costs US$10 billion, has zero residual value, is put into use immediately, and is depreciated straight-line over six years; the first customer contract lasts three years, and financing is assumed to mature in five years. At the end of year three, approximately US$5 billion of equipment carrying value remains undepreciated, and the financing has two years until maturity. If the customer does not renew, Oracle must subsequently find revenue, sell the equipment, or use other funds to pay the remaining obligations. This only illustrates the timing difference, not an actual matching of Oracle contracts or debt; US$5 billion is carrying value, not outstanding debt, and does not equal a loss. Actual profit or loss depends on cash recovered in the first three years, repayment of financing principal, later rents, costs, and equipment residual value.
The assessment should first track cash collected and still-enforceable customer commitments, then estimate rent, utilization, electricity, and maintenance costs over the remaining equipment life, and finally compare interest, maturing debt, data center rent, replacement, and exit costs that must be paid. Oracle has not disclosed enough project-level matched data to calculate a reliable cash recovery shortfall for each batch of GPUs; this is therefore a risk framework, not a conclusion that losses have occurred.
4. US$664 billion in RPO: first examine timing, then whether customers can pay
At the end of August 2026, Oracle's remaining performance obligations (RPO) were US$664 billion, up US$209 billion from US$455 billion a year earlier. The 10-Q expects approximately 13% to be recognized in the next 12 months, approximately 37% in months 13 through 36, approximately 34% in months 37 through 60, and the remainder later. Mechanically applying the company's proportions yields approximately US$86.3 billion over the next 12 months, but this is estimated recognition of existing period-end RPO, not next year's revenue guidance, much less cash already collected or distributable profit.10-Q performance obligations note
For a cloud contract to become shareholder returns, it must pass through at least five stages: obtaining hardware and power, delivering capacity as contracted, actual customer usage or performance, collection, and finally deduction of equipment depreciation, rent, operating and maintenance costs, and financing costs. Delay at any stage may lengthen recovery. RPO helps indicate visible demand, but cannot directly be treated as unconditional revenue; where cancellation, delay, and payment terms are not sufficiently disclosed, one cannot assume all orders can be canceled or that none can be canceled.
The company said it added more than US$30 billion in AI cloud contracts during the quarter and delivered an additional 850 MW of data center capacity. This supports the existence of new orders and actual delivery; but 850 MW is not the same as capacity sold, nor a measure of profitability. Investors must still examine when data centers enter service, when customers begin paying, and capacity utilization over the asset's entire life.Company release
This research focuses on large AI customers and OpenAI. Oracle has not provided enough information here to allocate US$664 billion in RPO among all individual customers or determine that any customer is about to default. Reasonable checks on large AI customers include whether their end-user payments and financing support commitments, how payment and prepayment terms are structured, and whether future capacity can be transferred to other customers. Prepayment can relieve some construction funding pressure, but also means Oracle must still provide future services; it is not free revenue.
Can large customer orders be canceled? Distinguish contractual rights from actual collection
There is currently no basis to reach a conclusion on contracts with OpenAI or other specific large customers. OpenAI's public statement confirms cooperation with Oracle to develop additional data center capacity, but does not disclose the complete minimum purchase, cancellation, penalties, delivery conditions, and payment protection needed here. Oracle's 10-Q also does not disclose these terms and RPO shares for each large customer. Thus, one cannot say all large orders can be canceled without cost, or that every one is absolutely noncancelable.
One available reference is Oracle's public US online cloud services standard agreement. The template generally states that orders are noncancelable and payments nonrefundable, but expressly allows exceptions in the agreement or order. It establishes only the terms of that standard template; it does not establish that OpenAI or other large AI customers use the same template or have no individual amendments. Minimum usage, default compensation, delivery conditions, and transfer rights in large contracts depend on the documents each party signed.
Different terms lead to different outcomes. If a customer can legally cancel, the unperformed portion may be removed from or adjusted in RPO, reducing future recognized revenue; if a contract has minimum commitments or termination fees, Oracle may have a right to recover, but conditions, offsets, and the customer's ability to pay still matter. Noncancelability is a legal right, not a guarantee that cash will be received. If a customer has financial difficulties, even with a contract, collection may be delayed, renegotiated, restructured, or subject to credit loss; there is insufficient information to determine that any particular customer is in those circumstances.
⚠️ Practical warning: Microsoft is not renewing some Oracle AI capacity. On November 6, 2026, Bloomberg cited people familiar with the matter in reporting that Microsoft was winding down an Oracle cloud rental arrangement dating to 2023 involving servers with older-generation GPUs; the report did not establish the full contract value. This is currently a media report concerning nonrenewal and gradual wind-down; it cannot establish early default or termination of all Microsoft contracts, nor allow an estimate of Oracle's revenue loss or impairment. The same report quoted Oracle's response: the company said demand for AI infrastructure remained strong, including for older-generation GPUs, and said it had stated in September that four-year GPU contracts were renewing at prices 20% above original contract prices. This is the company's statement; the report did not provide contract size and this research cannot independently verify it. It coexists with Microsoft's nonrenewal as a second signal, and it is not yet possible to judge which is more representative.Bloomberg report
This raises a direct test: if the same equipment still has carrying value but the original customer does not renew, Oracle must find other workloads to continue recovering its costs, and the speed and price of re-rental and electricity costs determine the impact. If other customers take over, the equipment may continue generating cash; if demand or prices are insufficient, recoverability must be assessed under the accounting rules applicable to the asset or asset group. This is a risk transmission mechanism, not a confirmed impairment.
The impact on Oracle also depends on the stage of investment. Capacity not yet purchased or committed may see reduced investment; GPUs already purchased can be offered to other customers, but redeployment takes time and may require price or specification discounts; signed lease and power commitments do not automatically disappear when customer demand weakens, and the ability to sublease, modify, or terminate them must be checked separately. If customers provide GPUs, Oracle may bear less equipment principal risk, but data center and service commitments remain; if Oracle funds them itself, customer loss affects equipment utilization, depreciation, and financing costs together.
Prepayments are not an all-purpose safeguard either. Depending on the contract, prepayments may be applied to future services, refundable under certain conditions, or affected by disputes and restructuring; without knowing the terms, one cannot assume the full amount can be retained. Oracle has disclosed that some large AI contracts involve customer prepayment for GPUs or customer-supplied hardware, but that does not imply all customers prepay or that all prepayments are nonrefundable.Oracle FY2026 Q4 release
The key questions are whether customer commitments are conditional, whether Oracle delivers on time, whether payment is secured, whether payments received can be retained, and whether capacity can be resold after exit. Oracle's 10-K warns that changes in key customers' ability to perform or demand may affect operating margins, cash flows, and recoverability of related long-lived assets; this is a disclosed risk, not evidence of cancellation by a particular customer.Oracle 10-K
5. Cash flow: prepayments support the quarter and cannot be deducted again as new revenue
Q1 operating cash flow of US$23.103 billion included an increase of US$11.363 billion in customer prepayments with a significant financing component within deferred revenue, approximately 49% of operating cash flow. Mechanically deducting it leaves approximately US$11.740 billion; this is only a sensitivity illustration, not recurring operating cash flow reported by the company, and does not imply that other working capital changes are repeatable. Prepayments are actual cash received in the period and their reality cannot be denied; but they arise from future service commitments and cannot be assumed to recur at the same scale every quarter.
Capital expenditures were US$28.499 billion and company-defined free cash flow after deducting them from quarterly operating cash flow was negative US$5.396 billion. If the US$11.363 billion prepayment above is also deducted in a sensitivity illustration, the mechanical difference is negative US$16.759 billion. This is not a replacement for GAAP free cash flow, but answers a specific question: how much larger would the quarterly funding gap be if future prepayments were not of the same scale and everything else remained exactly unchanged? In reality, revenue, collections, payment timing, and investment pace will all change, so the difference cannot be extrapolated as a full-year forecast.
The company separately reports US$17.966 billion in “net cash outlay for capital expenditures,” calculated as US$28.499 billion plus US$830 million of net repayment of short-term financing related to capital expenditures, less US$11.363 billion in customer prepayments. This measure helps explain the funding the company must raise itself, but prepayments are already included in operating cash flow. Deducting US$17.966 billion from the full US$23.103 billion CFO and calling it another FCF would double-count the same prepayment. Neither measure can omit future performance costs.Company schedules and definitions
6. Capital allocation: do external funding, interest, and dilution buy sufficient returns?
During the quarter Oracle completed an approximately US$20 billion at-the-market (ATM) equity program, generating US$19.909 billion in cash after issuance costs. Issuing shares gives the company more flexibility to invest, but reduces existing shareholders' share of the returns from new assets. Period-end common shares increased from approximately 2.880 billion at the end of the preceding quarter to 3.024 billion, approximately 5%; this is the change in total shares outstanding and includes issuance and employee plans, among other factors, and cannot be attributed entirely to the ATM program. If total company earnings rise but earnings and free cash flow per share do not keep pace with share growth, shareholders do not necessarily receive the same benefit.10-Q equity and cash flows
At the end of August, cash, cash equivalents, and marketable securities totaled US$37.077 billion; the company disclosed approximately US$125.0 billion of outstanding senior notes and other long-term borrowings. The difference of approximately US$87.9 billion is only a simplified difference between selected borrowings and liquid assets, not net debt encompassing leases and all liabilities. Quarterly interest expense rose from US$923 million to US$1.428 billion, approximately 55%. New investments must therefore cover both operating and funding costs; even if accounting operating income grows, interest and per-share cash must be monitored.10-Q fair value and interest expense
Off-balance-sheet financing and commitments not yet recognized: where are the risks?
There are significant lease commitments not yet recognized. Oracle's FY2027 Q1 note disclosed approximately US$288 billion in additional lease commitments as of August 31, 2026, almost all related to data centers. These leases had not commenced and were therefore not in the consolidated balance sheet or the maturity schedule for commenced leases at that time; they were expected to commence between FY2027 Q2 and FY2029, with terms of approximately 15 to 19 years. These future commitments are disclosed in the financial statement notes, not currently recognized debt, cash due within one year, or undisclosed liabilities; once the leases commence, they should be recognized under applicable accounting rules. They nevertheless constrain future flexibility: if customer contracts change, Oracle's ability to take on other workloads, sublease, or renegotiate will affect actual losses, and the entire commitment cannot be treated directly as a loss.Oracle Q1 10-Q lease note
By contrast, recognized lease liabilities were approximately US$34.621 billion in operating leases and US$9.185 billion in finance leases, totaling US$43.806 billion. Both are discounted carrying liabilities; US$288 billion is the commitment for leases not yet commenced, with different measurement, timing, and discounting bases. They cannot simply be added and called total debt on a consistent basis, nor can all operating leases be called off-balance-sheet.
The 10-Q also disclosed US$34.150 billion in unconditional purchase and other commitments extending beyond one year, mainly for cloud equipment components and data center power. These contractual commitments are listed separately from leases and should not directly be treated as funds already spent or liabilities already due; terms and potential overlap still require examination.Oracle Q1 10-Q other commitments
How does project financing transmit risk? An external developer or project company can raise financing to build a data center and then lease it to Oracle. The developer's borrowing is not necessarily Oracle's borrowing; but Oracle's rent, guarantees, and other contractual obligations may still create economic exposure. The existence of special-purpose entities, whether Oracle must consolidate them, guarantee amounts, and overlap with the US$288 billion commitment cannot be determined merely from third-party construction.
One example is Project Jupiter in New Mexico. Axios reported in September 2026 that Oracle gave developer Stack Infrastructure a force majeure notice; Oracle said the project remained on plan, while Blue Owl said the notice had not changed its financial commitment. The notice itself does not mean lease cancellation, a confirmed delay, blanket relief for Oracle, or a loss already incurred; it shows that power, construction, lease, and financing risks must be tracked individually against contracts.Axios project report
Overall assessment: Oracle has significant disclosed lease commitments not yet recognized at that time, as well as on-balance-sheet lease liabilities and other purchase obligations. If large customers do not renew, equipment rapidly loses value, and new leases commence in succession, revenue, equipment value, and fixed commitments may fall out of sync; however, available information is insufficient to calculate overlap or exact losses, or attribute all external project borrowing to Oracle as off-balance-sheet debt.
Retirement costs: include them in overall asset recovery rather than estimating losses in isolation
Exit costs include equipment removal, data erasure from storage media, transportation and recycling, and site restoration potentially required by leases; equipment or components still usable may offset some costs. Oracle's hardware recycling program covers Oracle-branded products returned from internal labs and data centers, and its description does not mention GPUs; Oracle also does not separately disclose OCI GPU retirement costs and residual values, so they cannot be assumed to be zero or estimated using peers' amounts.Oracle hardware recycling information
Peer CoreWeave recognized a US$120 million asset retirement obligation at the end of June 2026, mainly for expected removal of equipment associated with data center improvements; Equinix recognized US$228 million at the end of 2025, mainly for data center site restoration. These are different companies, dates, and asset scopes, and are not per-GPU disposal costs or dedicated cash set aside for this purpose.CoreWeave 10-Q; Equinix 10-K
If a contract or law creates a reasonably estimable retirement obligation, a company must assess recognition under applicable accounting rules; voluntary equipment replacement, operating disposal costs, and impairment cannot all be equated with retirement liabilities. Oracle did not separately list a related amount in the annual and quarterly reports reviewed here, which does not establish that there is no obligation or that one was omitted. Investment models should consider exit spending less recoverable value and avoid deducting a recognized liability and the same future cash flow twice.
7. The 2000 fiber-optic example: demand can materialize while capital returns still disappoint
Telecommunications and fiber-optic investment around 2000 shows that growing technology demand does not guarantee good returns for those who build first. Global Crossing, for example, built a global fiber-optic network and originally planned to fund expansion through capacity sales or asset sales; the telecom market subsequently weakened, demand fell, and access to capital markets became difficult, and the company and some subsidiaries filed for bankruptcy protection in 2002. This illustrates possible mismatches in the timing of demand, capital investment, and financing. Subsequent growth in internet and data traffic does not automatically compensate every asset holder or shareholder from that period. The historical lesson concerns investment price, debt structure, cash recovery timing, and who owns the assets, not merely whether the technology succeeds.Global Crossing restructuring document filed with the SEC
Oracle differs from the fiber-optic companies of that era: it has mature software and enterprise customers, ongoing operating revenue, some customer prepayments, and an ability to raise funds through share issuance; equipment lives, leases, and contract structures for AI data centers also differ. History therefore cannot predict Oracle's failure. The questions to draw from it are: if ultimate demand grows more slowly than added capacity, who bears fixed costs first? If full utilization takes longer, must existing shareholders accept more dilution or lower returns?
8. Cross-examination of the bull and bear cases
Bull: With IaaS up 121% year over year, can AI demand still be called just a story? No. Quarterly revenue and capacity delivery have already demonstrated that some demand has entered actual results. The bear case really asks whether this revenue, after asset replacement, power, rent, and interest, is enough to cover invested capital; public information is currently insufficient to calculate complete incremental ROIC by project.
Bear: Do negative free cash flow and large leases mean Oracle has already run out of funding? That cannot be inferred. The company has cash flow from existing businesses, customer prepayments, completed share issuance, and cash assets at period-end. The risk is increased financing dependence and concentrated future fixed obligations, not an immediate default demonstrated by this article.
Bull: Don't large customer prepayments or customer-supplied GPUs transfer the risk away? They can reduce or change Oracle's initial funding needs and are valuable contractual arrangements. But customer prepayments correspond to subsequent services, and Oracle must still maintain power, networking, data centers, and service levels. The specific allocation of returns and risks depends on each contract's terms; one term cannot be generalized to all contracts.
Bear: As models become more efficient, could Oracle's GPUs sit idle? They could, but efficiency could also lower service costs and encourage more enterprises to use them. Equipment efficiency and demand elasticity will jointly determine the outcome. Oracle has not fully disclosed utilization by data center, GPU service gross margins, or project payback periods, so it is not currently possible to quantify which force is stronger.
9. What should investors track in the next earnings report?
Track recovery through the four clocks. On the customer side, examine contract terms, renewals, minimum commitments, and collections; on the equipment side, examine utilization, realized rents, power costs, depreciation, and impairment for different GPU generations; on the funding side, examine debt maturities, interest, commenced and uncommenced leases, purchase commitments, and changes in share count. Then deduct cash capital expenditures from operating cash flow and continuously monitor changes in prepayments and free cash flow per share. Where the company does not provide a breakdown, explicitly identify an information gap rather than substitute market quotes for Oracle's actual prices.
In its Q1 release, management forecast FY2027 Q2 revenue growth of 30% to 34% year over year and cloud revenue growth in US dollars of 65% to 71%; full-year revenue of at least US$90 billion and non-GAAP EPS of US$8.10. These were the company's forecasts at the time, not realized performance.Company guidance
Five things should be checked continuously: first, whether IaaS growth comes with timely capacity launches rather than dependence on a few large customers; second, whether the timing, collection, and performance of customer prepayments and RPO align; third, whether incremental equipment spending, depreciation, leases, and interest are absorbed by additional gross profit; fourth, whether operating cash flow still requires substantial prepayment support; fifth, whether free cash flow per share can improve after changes in borrowing and share count.
The positive scenario is higher utilization of new capacity, continued customer payment, and operating cash flow growing with revenue and exceeding maintenance and expansion needs; the neutral scenario is continued strong revenue but a longer investment cycle and delayed cash recovery per share; the negative scenario is disrupted delivery or customer payment while fixed lease and equipment costs continue, further increasing the need for borrowing or share issuance. Which scenario occurs must be verified through subsequent financial statements and contract evidence.
Q: How much of Oracle's cloud growth comes from existing database customers?
It cannot currently be determined. Q1 IaaS revenue was approximately US$7.4 billion, SaaS approximately US$4.2 billion, and traditional software revenue US$5.55 billion, down 3% year over year; the company did not break out how much incremental cloud revenue came from existing database customers and how much from large AI contracts. There is insufficient evidence to interpret all IaaS growth as a natural extension of the old moat; the more cautious view is that both sources coexist in unknown proportions.
Q: Will Oracle's US$664 billion RPO become cash?
It will not automatically become cash and is not equivalent to profit. RPO is contracted revenue not yet performed; the 10-Q expects approximately 13%, or approximately US$86.3 billion, to be recognized over the next 12 months. This is estimated recognition, not next year's revenue guidance. For contracts to become shareholder cash, Oracle must obtain hardware and power, deliver as contracted, have customers actually use services, collect payment, and then deduct depreciation, rent, operating and maintenance costs, and financing costs.
Q: Why is Oracle's free cash flow negative, and has it run out of money?
Negative free cash flow results from expansion outpacing operating cash flow, and funding failure cannot currently be inferred from it. FY2027 Q1 operating cash flow was US$23.103 billion and capital expenditures US$28.499 billion, with company-defined free cash flow negative US$5.396 billion; full-year FY2026 was negative US$23.686 billion. The company has cash flow from existing businesses, customer prepayments, and approximately US$20 billion in share issuance to support it; the risks are increased financing dependence and concentrated future fixed obligations.
Q: Are customer prepayments incremental cash flow for Oracle?
They are actual cash receipts, but cannot be treated as revenue that will recur every quarter. Of Q1 operating cash flow of US$23.103 billion, the increase in customer prepayments with a significant financing component was US$11.363 billion, approximately 49%; approximately US$11.740 billion remains after deduction, but this is only a sensitivity illustration, not recurring cash flow reported by the company. The prepayment is already included in operating cash flow, so “net cash outlay for capital expenditures” of US$17.966 billion cannot be deducted again from US$23.103 billion without double-counting.
Q: Is six-year GPU depreciation reasonable?
There is currently no universal answer. Oracle uses the straight-line method and estimates six-year useful lives for its principal servers and network equipment, but does not individually disclose actual GPU lives; Microsoft lists two to six years, while Amazon lists five to six years and shortened some equipment from six years to five in 2025. Oracle extended lives from five to six years in FY2025, reducing operating expenses by approximately US$733 million and increasing diluted EPS by approximately US$0.20. What matters is equipment utilization, realized rents, electricity and maintenance costs, and whether impairment is reflected promptly.
Q: Does Microsoft's nonrenewal of Oracle AI capacity mean demand is weakening?
That cannot currently be inferred. A November 6, 2026 media report said Microsoft was winding down a rental arrangement dating to 2023 using older-generation GPUs; the report did not establish contract size and does not mean default or impairment. Oracle responded that demand for AI infrastructure remained strong and said it had stated in September that four-year GPU contracts were renewing at 20% above original contract prices; this is the company's statement and cannot be independently verified. The two signals coexist, and it is not yet possible to judge which is more representative.
Q: Is the US$288 billion lease commitment Oracle's debt?
It is not recognized debt or cash due within one year. It is an additional lease commitment as of August 31, 2026, not yet commenced and almost entirely related to data centers, expected to commence between FY2027 Q2 and FY2029 with terms of approximately 15 to 19 years. Recognized lease liabilities are approximately US$34.621 billion in operating leases and US$9.185 billion in finance leases, totaling US$43.806 billion; the measures and discounting bases differ and cannot be added directly. Recognition under applicable accounting rules occurs after leases commence.
Q: How does share issuance affect Oracle shareholders?
Shareholders' proportionate share of returns from new assets falls unless earnings and free cash flow per share keep pace. During the quarter Oracle completed an approximately US$20 billion at-the-market (ATM) equity program, bringing in US$19.909 billion after costs, and period-end common shares increased from approximately 2.880 billion to 3.024 billion, approximately 5%; this is the change in total shares, includes employee plans and other factors, and cannot all be attributed to issuance.
This article is based on complete financial reports and related information obtained before November 9, 2026. It does not include consensus market estimates or valuation at the latest share price and does not infer whether current market pricing is reasonable; its core financial information relies on company releases and SEC filings. This article is research based on public information and investment education, and does not constitute investment advice.
ProfitVision LAB|Independent Research
