Oracle’s AI infrastructure expansion is supposed to print money. In the regulatory filing it published late last month, the company essentially admits it might not, and it breaks down why in painstaking detail.

The centerpiece of the risk is Oracle Cloud Infrastructure, or OCI. To grow that business, Oracle wrote that it must incur “significant capital and operating expenditures” to increase existing data center capacity and establish data centers in new geographic locations. Those investments come with long-term commitments for infrastructure and capacity, which means Oracle is not just buying compute. It is signing up to build and lease capacity that has to be used, paid for, and ultimately renewed. When customers do not renew their contracts, Oracle warned: it “may be unable to re-lease, repurpose or assign such capacity on acceptable terms, if at all.”

This is happening in a very specific AI supply-chain reality. Unlike the big three cloud providers, Oracle prefers to lease data center capacity from partners like Crusoe rather than build it all itself. That hybrid model is often framed as agility. In the filing, it reads more like exposure: if demand is wrong, Oracle may still be on the hook.

Oracle’s AI bet is not theoretical. In early 2025, Oracle joined OpenAI, SoftBank, and MGX to put its name on the so-called Stargate initiative, an ambitious project described as paving the planet with “half a trillion dollars” worth of “bit barns.” Oracle also signed a long-term agreement with OpenAI under which it would provide $300 billion of capacity over five years. The database provider is also described as managing the model dev’s flagship facility in Abilene, Texas.

Oracle also says it has remaining performance obligations of about $155 billion from other customers. In normal cloud reporting, that would look like a cushion. Here, it functions more like a clock with multiple failure points. Oracle laid out the dilemma in plain English terms: if it underestimates demand, it could lose customers to competing infrastructure providers. If it overestimates demand, or if key customers can’t make payments, Oracle could end up footing the bill for leased datacenter capacity. Even the “success” scenario depends on customers continuing to pay, staying solvent, and choosing to renew.

Then there is the part of the AI buildout that CEOs rarely put on the slides: electricity, hardware, and regulatory friction. Oracle says it has faced, and may continue to face, challenges securing reliable and cost-effective power sources for data centers, with the constraint described as global due to increased demand for and limited availability of energy to power AI compute. Power prices can be volatile, including due to extreme weather events and market structure in certain regions. Oracle warned that increases in energy costs can adversely affect margins, particularly where customer pricing is fixed or committed.

The filing also calls out a whole chain of operational dependencies. Oracle’s expansion depends on access to suitable, permitted build sites, reliable and predictable power sources, networking hardware, and server availability, including graphics processing units, memory devices, and other critical components. Data centers in geographies it relies on may become unavailable on commercially reasonable terms or at all. It also flags government-imposed limits or moratoria on data center construction, plus a wide list of execution risks tied to design, engineering, permitting, construction, utility interconnection, equipment delivery, and contractor performance.

The regulatory and physical world is not an abstraction either. Oracle lists laws and regulations around land use and zoning, environmental permitting, energy usage, grid reliability, greenhouse gas emissions, water usage, building codes, health and safety, tax incentives, and data localization. That is the paperwork version of “everything can go wrong,” but it is also the point: Oracle is committing capital and capacity into an environment where constraints can shift.

And Oracle is committing anyway. The company warned investors that it expects to continue to invest significant resources to build and support its AI products in line with its growth strategy, warning it could fall behind if it does not invest. During its Q4 earnings call last month, Oracle said it planned to spend $70 billion on capital expenditures during its 2027 fiscal year, up from around $55 billion spent during its 2026 fiscal year. To fund that, Oracle hopes to raise around $40 billion in debt and equity in 2027, on top of $18 billion in debt it raised back in September.

The market is not ignoring the math. The filing and the spending plan sit alongside a stock that is down more than 40 percent in the last month. Meanwhile, Oracle’s OpenAI deal is reportedly expected to contribute up to $30 billion in revenues annually, with revenues expected as early as next year, but the filing underscores that OpenAI still hasn’t managed to turn a profit. Oracle’s ability to get paid depends on its customers continuing to raise capital and meeting their commitments.

So what does this mean beyond Oracle? If you are an operator or investor watching AI infrastructure, this is a live stress test for an emerging model: building or leasing massive compute capacity before demand is perfectly visible, while fixed pricing, power constraints, permitting timelines, and customer renewal risk can stack. Oracle’s filing reads like a corporate legal document, but the subtext is strategic. It is a reminder that the AI data center race is not only about GPUs. It is also about who can absorb the downside when the demand curve, the power bill, or customer cash flows do not line up neatly.