Has Oracle spent too much on AI?

Why is Oracle spending so much money on AI?

Oracle is spending extraordinary amounts on AI because it wants to become one of the world's biggest suppliers of computing power, and its customers are already signing contracts worth hundreds of billions of dollars.

The company is building massive data centers filled with GPUs for customers including OpenAI, Meta and xAI. In fiscal 2026, Oracle spent $55.7 billion on capital expenditures, compared with $21.2 billion the previous year and just $6.9 billion in fiscal 2024. That is an eightfold increase in two years.

The spending accelerated further in the first quarter of fiscal 2027, reaching $28.5 billion in just three months. For perspective, Oracle invested more in that single quarter than it spent during the entire fiscal year 2025. Numbers like $28.5 billion in one quarter make it worth knowing how big the market really is today; we cover that in our report on AI infrastructure.

The demand is real. Oracle Cloud Infrastructure revenue grew 121% year over year to $7.4 billion in the latest reported quarter, while its remaining performance obligations, essentially revenue promised under existing contracts, reached $664 billion.

But spending $28.5 billion in a quarter when the entire company generated $19.3 billion in revenue is unusual, even for the technology industry. Oracle is making a huge bet that tomorrow's AI computing demand will justify today's construction and equipment costs.

The question is whether those contracts will eventually produce enough cash to cover the infrastructure, financing and hardware replacement costs. For now, Oracle has proved that customers want its computing capacity. It has much more to prove about the returns shareholders will earn from supplying it.

Has Oracle's AI spending grown faster than its business?

Oracle's AI spending has grown dramatically faster than its business, with capital expenditures increasing more than four times as fast as operating cash flow over the past two fiscal years.

According to Oracle's annual filings, capital expenditures jumped from $6.9 billion in fiscal 2024 to $55.7 billion in fiscal 2026. Over the same period, operating cash flow increased from $18.7 billion to $32.0 billion. Spending rose approximately 711%, compared with 71% growth in cash generated by operations.

The contrast becomes even sharper when we look at the latest quarter. Oracle recorded $28.5 billion in capital expenditures against $19.3 billion in revenue, meaning it invested approximately $1.48 for every dollar of sales.

Most of the equipment and infrastructure will generate revenue over future periods, but the ratio shows just how far ahead of its current business Oracle is building.

Historically, Oracle operated a much less capital-intensive business. Its database and enterprise software products generated recurring revenue without requiring anything close to today's infrastructure investment.

Fiscal period	Capital expenditures	Operating cash flow	Free cash flow
FY2023	$8.7B	$17.2B	+$8.5B
FY2024	$6.9B	$18.7B	+$11.8B
FY2025	$21.2B	$20.8B	-$0.4B
FY2026	$55.7B	$32.0B	-$23.7B
FY2027 Q1	$28.5B	$23.1B	-$5.4B
FY2024–FY2026 change	+711%	+71%	-$35.5B
FY2027 Q1 capex/revenue	148%	-	-

Is Oracle actually making money from its AI investments?

Oracle's AI infrastructure business is already generating billions of dollars in additional revenue, although we still cannot tell how profitable the latest data-center investments will be over their full lifetimes.

In the first quarter of fiscal 2027, Oracle Cloud Infrastructure revenue reached $7.4 billion, up from approximately $3.3 billion a year earlier. That represents about $4.1 billion in additional quarterly sales.

The increase is especially impressive when compared with the rest of Oracle. Total company revenue grew 30% to $19.3 billion, while cloud applications revenue increased just 10% to $4.2 billion. Software revenue actually declined 3%.

Infrastructure therefore accounted for more than 90% of Oracle's year-over-year revenue increase during the quarter. Its share of total company revenue climbed from around 22% to 38% in twelve months.

Oracle also reported delivering more than 300,000 additional GPUs to AI cloud customers after the end of fiscal 2026, along with 850 megawatts of additional data-center capacity during its first fiscal quarter.

Oracle reported $6.7 billion in company-wide GAAP operating income for the quarter, up 57%, but it does not separately disclose the full lifecycle profitability of its newest AI data centers. Total infrastructure revenue also includes traditional cloud workloads alongside AI.

Does Oracle really have $664 billion in future revenue guaranteed?

Oracle's $664 billion backlog represents an enormous amount of contracted future revenue, but treating the entire figure as guaranteed cash would be a serious mistake.

At the end of the first quarter of fiscal 2027, Oracle reported $664 billion in remaining performance obligations, or RPO. The figure was $455 billion a year earlier and $638 billion at the end of fiscal 2026.

That means Oracle added $209 billion to its backlog over twelve months, an increase of approximately 46%. Its latest results also showed more than $30 billion of additional AI cloud contracts booked during the quarter. A backlog that grew by $209 billion in a year is the sort of shift we go back over in our report covering AI infrastructure.

To understand the size, Oracle generated approximately $67 billion in total revenue during fiscal 2026. Its contracted backlog is therefore equivalent to almost ten years of sales at that historical revenue level.

Of course, those contracts will be delivered over different periods, and the backlog includes software and other cloud services alongside AI infrastructure.

The bigger issue is that RPO measures contracted revenue before operating costs. A $10 billion computing contract could be extremely profitable or only marginally profitable, depending on the cost of GPUs, electricity, data-center construction and financing.

Contract terms also matter. Oracle's public disclosures do not provide everything needed to assess customer cancellation rights, payment schedules or profitability for individual agreements.

Is Oracle borrowing too much money and burning too much cash on AI?

Oracle is becoming dangerously reliant on external financing, with its AI expansion requiring far more capital than its existing business can generate internally.

In fiscal 2026, the company generated approximately $32 billion in operating cash flow but spent $55.7 billion on capital expenditures. That left a $23.7 billion gap before considering other financing needs.

Oracle raised $43 billion through debt financing and another $5 billion through equity financing during that fiscal year. Management then outlined roughly $40 billion in additional financing for fiscal 2027, including a $20 billion common-stock issuance program completed in its first quarter.

This represents a major departure from Oracle's traditional financial model. In fiscal 2024, operating cash flow covered capital expenditures almost three times over. The company could invest in its business, pay dividends and still have cash available for debt reduction or other purposes.

The change is also visible in free cash flow. Oracle generated positive free cash flow of $11.8 billion in fiscal 2024, approximately negative $394 million in fiscal 2025 and negative $23.7 billion in fiscal 2026.

The latest quarter brought another $5.4 billion deficit, even though Oracle reported record operating cash flow of $23.1 billion.

That unusually strong quarterly cash inflow deserves closer examination. Oracle's financial statements included $11.36 billion in customer prepayments with significant financing components related to capital expenditures. These payments help fund infrastructure before Oracle recognizes all the revenue associated with delivering the contracted services.

Oracle also introduced a measure called net cash outlay for capital expenditures. For the first quarter of fiscal 2027, this stood at approximately $18.0 billion, compared with $28.5 billion in reported capital expenditures.

The difference reflects specific customer and financing arrangements rather than a reduction in the physical investment being made.

Borrowing avoids shareholder dilution but adds interest payments and refinancing obligations. Issuing shares reduces the need for debt while spreading future earnings across more shareholders.

The financing risks are already attracting attention. Recent Barron's reporting highlighted Oracle's BBB- credit rating from S&P Global Ratings, the lowest investment-grade category. A downgrade could increase borrowing costs and make some institutional investors less willing or able to hold its debt.

Oracle's conventional free cash flow has now been negative for six consecutive quarters, according to recent financial reporting. Customer financing helps, but the company is still taking on substantial financial obligations before its newest infrastructure has produced the expected returns.

Are Oracle's AI customers paying for the GPUs themselves?

Oracle's largest AI customers are now helping finance tens of billions of dollars in hardware, substantially reducing the amount Oracle must borrow or raise from shareholders.

In its fiscal 2026 earnings announcement, Oracle disclosed that the prepaid and customer-supplied hardware portions of large AI contracts had reached approximately $75 billion. Customers prepaying for their own GPUs changes how a cloud works; we compare which business models are winning in our report on AI infrastructure.

Under some agreements, customers prepay Oracle for the GPUs required to deliver computing capacity. Under others, customers purchase the GPUs themselves and supply them to Oracle.

This arrangement is particularly important because accelerators are among the most expensive components of AI infrastructure.

For perspective, the $75 billion figure exceeds Oracle's entire $55.7 billion capital expenditure total for fiscal 2026. These amounts cover different periods and accounting categories, so they cannot simply be subtracted from one another. Still, the comparison shows that customer participation is financially substantial.

Oracle's first-quarter results offered further evidence of this approach. Despite announcing more than $30 billion in additional AI cloud contracts, management said the structure of those agreements would not increase its existing capital-raising plans.

The arrangement also changes who bears some of the risks associated with GPU depreciation and replacement. Customer-owned equipment leaves less direct hardware investment on Oracle's balance sheet.

There are limits to the protection. Customer prepayments come with future service obligations, while Oracle still has to provide facilities, networking, cooling and other infrastructure.

Even so, we should give Oracle credit for negotiating contracts that push some of the industry's enormous equipment costs back toward customers.

Has Oracle become too dependent on OpenAI?

Oracle is now heavily exposed to OpenAI, and that concentration is one of the biggest weaknesses in its AI investment strategy.

In September 2025, The Wall Street Journal reported that OpenAI had agreed to purchase approximately $300 billion in computing capacity from Oracle over roughly five years, beginning in 2027.

A contract of that size would be extraordinary for almost any infrastructure company. Compared directly with Oracle's latest $664 billion backlog, it represents roughly 45%.

That is an indication of scale rather than a verified current customer-concentration ratio. Oracle does not provide a complete public breakdown of its backlog by customer, and the value of a reported multiyear agreement does not necessarily equal its remaining accounting obligations.

The concentration becomes even clearer when we compare the reported contract with Oracle's existing business. A simple $300 billion divided by five years gives an average of $60 billion annually, nearly equivalent to Oracle's entire fiscal 2026 revenue of approximately $67 billion.

Actual payments and revenue recognition could be distributed very differently across the contract period.

Oracle does serve other major AI customers. Its financing announcements have named Meta, xAI, Nvidia, AMD and TikTok, among others.

But those relationships do not eliminate the scale of the OpenAI exposure. A major reduction, delay or restructuring of OpenAI's computing requirements could affect Oracle's revenue expectations, infrastructure plans and financing needs.

Oracle has won one of the industry's most valuable customer relationships. It has also tied a large part of its future growth to that customer's financial success. That's a pretty uncomfortable position for a company committing this much capital.

Can OpenAI actually afford its enormous Oracle contract?

OpenAI has become large enough to make extraordinary computing commitments, but its ability to finance them over several years remains a genuine concern for Oracle.

The reported $300 billion agreement with Oracle is part of a much broader infrastructure expansion involving Microsoft, Amazon, Nvidia and other partners.

OpenAI's computing requirements have grown quickly as its models have become more capable and its products have attracted more users. Yet the infrastructure commitments extend far beyond the scale of its historical revenue.

The latest reporting makes the situation more uncertain. The Wall Street Journal recently described setbacks in the Stargate initiative, including difficulties involving OpenAI's original negotiations with SoftBank and Oracle. OpenAI has subsequently pursued separate computing arrangements with other providers.

The company has also experienced changes in its infrastructure leadership. Sachin Katti, a senior executive responsible for computing capacity, was reported to be leaving after playing a major role in expanding OpenAI's infrastructure partnerships.

Financial Times reporting has meanwhile described broader problems across the US AI data-center boom, including financing difficulties, delayed projects and increasing questions about the industry's long-term spending plans.

None of this establishes that OpenAI will fail to meet its contractual obligations. The company continues to attract substantial investor interest and has pursued additional financing.

What concerns us is the difference between signing a multiyear infrastructure agreement and producing enough cash to support the payments.

Oracle's exposure therefore depends on OpenAI's ability to keep growing revenue, secure financing and maintain strong demand for the computing capacity it has committed to purchase.

Will Oracle's AI data centers actually be profitable?

Oracle expects its AI computing contracts to produce gross margins of around 30% to 40%, but those figures still leave important questions about the final returns on its infrastructure investments.

During its fiscal 2026 earnings discussions, management maintained that AI accelerator workloads could generate gross margins within that range over the lifetime of customer contracts.

The company also expects to sell additional cloud services alongside GPU computing. In March 2026, co-CEO Clay Magouyrk explained that storage, networking and general-purpose computing could account for another 10% to 20% of customer spending, often with higher margins.

That is commercially attractive. A customer running large AI training or inference workloads also needs storage, networking and related infrastructure. Oracle can potentially earn additional revenue from the same relationship without making proportionally identical investments in GPU hardware.

But a 35% gross margin does not automatically translate into a strong investment return.

Suppose a group of contracts generates $10 billion in revenue at a 35% gross margin. That produces $3.5 billion in gross profit over the contract period. We would still need to consider the capital required to provide the services, financing costs, taxes, corporate expenses and the timing of cash payments.

A five-year contract also has different economics from a ten-year agreement, especially when the underlying equipment becomes outdated quickly.

Oracle has disclosed enough to suggest that AI computing can be profitable. It has not disclosed enough for us to calculate the return on invested capital across its newest generation of data centers.

And that's the number we'd really like to see, given the unprecedented spending involved.

Could Nvidia GPUs become obsolete before Oracle earns its money back?

Oracle faces a real risk that its expensive AI accelerators lose economic value faster than expected, particularly as Nvidia introduces more powerful and efficient chips.

The pace of hardware development has been remarkable. Nvidia moved from Hopper to Blackwell and then toward Vera Rubin, bringing substantial improvements in computing performance, memory systems and efficiency. Hardware that keeps jumping a generation is one of the hard problems we examine in our report on AI infrastructure.

Newer GPUs do not automatically make older equipment useless. Many workloads can continue running effectively on previous-generation accelerators, especially when the customer values available capacity more than having the latest hardware.

Nevertheless, the economics can change quickly. If newer GPUs deliver substantially more computing power per watt, customers may eventually demand lower prices for older equipment.

A long-term contract with predictable payments can protect Oracle against some of this pressure. The amount of protection depends on whether Oracle must refresh the hardware, maintain specific performance levels or absorb changes in operating costs.

Customer-supplied GPUs also reduce Oracle's direct exposure to depreciation. The remaining uncertainty is substantial because Oracle does not publicly disclose detailed hardware replacement terms for its largest AI agreements.

Is Oracle catching up with Amazon, Microsoft and Google in AI cloud?

Oracle is growing much faster than the largest cloud providers in infrastructure percentage terms, but it remains far smaller than AWS, Microsoft Azure and Google Cloud across the broader cloud market.

Oracle Cloud Infrastructure revenue grew 121% year over year in its latest reported quarter. That is exceptional growth for a business already generating $7.4 billion in quarterly sales.

However, the comparison becomes more interesting when we look at what Oracle is actually selling.

AWS, Azure and Google Cloud have extensive platforms covering application hosting, databases, analytics, networking, security and AI services. Their customers include millions of businesses and developers running a wide range of workloads.

Oracle's recent acceleration is particularly tied to large infrastructure deployments. Its strategy is strengthened by existing enterprise database relationships, including the ability to offer Oracle database services through competing cloud platforms.

That creates a somewhat unusual competitive position. Oracle can compete with Microsoft and Google for AI computing contracts while also working with their cloud ecosystems to serve database customers.

The company's recent growth demonstrates that large AI developers are willing to buy enormous amounts of computing capacity from Oracle. Its ability to deliver 850 additional megawatts of data-center capacity in a single quarter reinforces that point.

There is much less evidence that Oracle is replacing AWS, Azure or Google Cloud as the default infrastructure platform for ordinary businesses. Oracle's clearest opportunity is large-scale contracted computing capacity, where customers increasingly care about price, available hardware and delivery schedules.

Is Oracle taking more risk than Microsoft, Amazon and Google?

Oracle is taking greater financing risk relative to its existing cash generation than the biggest cloud companies, even though all of them are spending heavily on AI.

Microsoft, Amazon, Alphabet and Meta have also committed enormous amounts of capital to data centers, accelerators and supporting infrastructure. Oracle is not the only giant spending at this scale, so we look at who is ahead and who is falling behind in our report covering AI infrastructure.

The difference lies in their ability to finance that spending from established businesses.

Microsoft benefits from Azure, Microsoft 365 and enterprise software. Alphabet generates substantial cash from Google Search advertising. Amazon has AWS, retail and advertising, while Meta relies on its huge social advertising business.

Oracle's database and enterprise software operations are profitable, but its fiscal 2026 operating cash flow of approximately $32 billion was well below its $55.7 billion capital spending.

That financing gap creates pressure that the largest hyperscalers are generally better positioned to absorb.

Oracle does have an advantage over younger GPU cloud providers such as CoreWeave and Nebius: a longstanding enterprise customer base, recurring software revenue and decades of experience running business-critical technology.

Still, the current expansion increasingly resembles the economics of a capital-intensive infrastructure provider rather than those of a traditional enterprise software company.

Oracle's competitive ambition is credible. Its balance sheet simply gives it less room for expensive mistakes than the industry's largest cash-generating technology companies.

Company	Main AI infrastructure business	Existing financial support	Main investment risk
Oracle	GPU cloud and AI data centers	Database and enterprise software	Debt, cash flow and customer concentration
Microsoft	Azure and AI infrastructure	Azure, Microsoft 365 and enterprise software	Returns on enormous AI investment
Amazon	AWS cloud and AI accelerators	AWS, retail and advertising	Capital intensity and computing efficiency
Alphabet	Google Cloud, TPUs and AI services	Google Search and advertising	AI costs versus monetization
Meta	AI infrastructure for its own products	Facebook and Instagram advertising	Whether AI improves advertising economics
CoreWeave	GPU cloud infrastructure	Contracted computing revenue and financing	Leverage, customer concentration and hardware costs
Nebius	AI cloud and data centers	Infrastructure contracts and external funding	Financing and delivering rapid expansion

Could electricity shortages and delayed data centers hurt Oracle?

Power shortages and construction delays could seriously undermine Oracle's AI investment returns by pushing revenue into later periods while financing and development costs continue accumulating.

Large AI data centers need enormous amounts of electricity, cooling equipment and grid infrastructure. These requirements are becoming harder to secure as technology companies compete for capacity in the same regions.

According to recent Financial Times reporting, at least 47 major US data-center projects have been stalled or canceled this year, while more than 1,000 communities have proposed restrictions on new developments.

The reporting also cited estimates suggesting that only around 35% of planned data-center projects may be feasible.

Those figures describe the wider US market rather than Oracle's own construction pipeline. Still, Oracle is building within the same constrained infrastructure environment.

The company demonstrated considerable execution capacity when it reported delivering 850 megawatts of additional data-center capacity during its fiscal 2027 first quarter.

But delivering one wave of projects successfully does not guarantee that every subsequent site can secure power connections, suitable land and construction resources on schedule.

Delays become expensive because infrastructure costs often begin accumulating before the first customer payment associated with operating capacity. GPU deliveries, financing arrangements and contractual commitments can also become more complicated when construction falls behind.

What would have to happen for Oracle's AI spending to pay off?

Oracle's AI investments can pay off if its completed data centers generate enough cash to cover construction, financing and hardware replacement costs, while reducing the need for continuous new borrowing.

We can put a number on the challenge using the company's fiscal 2026 results.

Oracle generated approximately $32 billion in operating cash flow while spending $55.7 billion on capital expenditures. At the same spending level, operating cash flow would need to increase by $23.7 billion, roughly 74%, to bring conventional free cash flow back to zero.

That calculation assumes spending stays constant and does not capture all the effects of customer financing.

The good news is that cloud infrastructure revenue is growing extremely quickly. The company also has hundreds of billions in future contractual commitments, and its customers have shown a willingness to help finance expensive hardware.

What we want to see next is more revenue coming from completed facilities without another equally dramatic increase in capital requirements.

That might happen as data centers become operational, customers begin making regular payments and Oracle spreads fixed infrastructure costs across larger computing workloads.

Oracle could also reduce its own capital requirements through greater use of customer-owned GPUs, prepayments and phased construction. Future contracts must offer enough expected profit to justify whatever capital Oracle still needs to commit.

What numbers would prove Oracle has spent too much on AI?

The clearest evidence that Oracle has overspent would be continued growth in AI contracts and revenue without a matching improvement in free cash flow, debt levels and returns on completed infrastructure.

We already know that demand is substantial. The more revealing numbers now concern how efficiently Oracle can turn that demand into cash.

Consider the relationship between capital spending and operating cash flow. Oracle's operating cash flow rose 54% in fiscal 2026, a strong result by ordinary standards. Yet capital expenditures rose 162%.

That difference is difficult to sustain indefinitely without increasing debt, issuing more shares or relying on customer financing.

We should also watch whether Oracle's contracted backlog continues converting into actual revenue at the expected pace. A large backlog becomes less reassuring if construction delays or customer financing problems prevent timely delivery.

Finally, Oracle's realized margins matter more than management's expected margins. The company needs to demonstrate that the large computing contracts generate attractive profits after equipment costs and financing requirements.

Indicator	Latest reference point	What we should watch
Cloud infrastructure revenue	$7.4B in FY2027 Q1	Growth from completed projects
Infrastructure revenue growth	+121% year over year	Whether strong growth continues at larger scale
Contracted future revenue	$664B	Conversion into recognized revenue
Annual capital expenditures	$55.7B in FY2026	Whether spending begins to moderate
Quarterly capital expenditures	$28.5B in FY2027 Q1	Whether the company keeps accelerating investment
Annual operating cash flow	$32.0B in FY2026	Growth relative to capital spending
Annual free cash flow	-$23.7B in FY2026	Movement toward sustained positive cash flow
Customer-funded hardware	$75B disclosed in FY2026	How much future equipment customers finance
Expected AI gross margin	30–40%	Actual margins and investment returns
Credit rating	S&P BBB-	Whether Oracle preserves investment-grade status

Has Oracle spent too much on AI?

Oracle has probably taken on more financial risk than it needed to, but calling its AI spending an outright commercial mistake would be premature.

The numbers are extreme. Capital expenditures rose eightfold in two years, free cash flow fell deeply into negative territory, and Oracle has become much more dependent on debt, equity issuance and customer financing.

The company has also secured something competitors would gladly pay for: enormous multiyear computing contracts, exceptional infrastructure revenue growth and a strong position with some of the world's largest AI developers. Big contracts with the largest AI developers do not guarantee a win, so we weigh the likely future winners in our AI infrastructure market report.

That makes Oracle different from a company building empty data centers in the hope that customers eventually appear. Its commercial demand is already substantial.

Our main concern is the relationship between that demand and the capital required to serve it. Oracle is committing money faster than it can generate cash, while depending heavily on a small number of exceptionally large customers, particularly OpenAI.

Recent concerns about AI infrastructure financing, weaker investor appetite for some data-center debt and delays across the Stargate ecosystem make that exposure more important than it looked when the contracts were first announced.

We would therefore describe Oracle's strategy as commercially promising but financially overextended. The company may ultimately generate excellent returns, especially if customer-funded hardware reduces its direct capital burden and its expected gross margins hold up.

For now, though, Oracle's shareholders are bearing substantial risks before the returns have been demonstrated.

The company has proved that customers are willing to sign enormous AI computing contracts. The remaining challenge is making those contracts profitable enough to justify what Oracle has already committed.

That is where we think Oracle has gone too far: the opportunity is real, but management has allowed the scale of investment to run ahead of the financial evidence that would normally justify it.