Logo
FrontierNews.ai

Oracle's $18 Billion AI Gamble: Why Banks Can't Sell the Stargate Debt

Oracle's ambitious $18 billion construction loan for OpenAI's Stargate project in New Mexico is stuck with underwriting banks because institutional investors are wary of the deal, revealing deeper concerns about Oracle's creditworthiness as it pivots toward AI infrastructure financing. The loans, which were supposed to be distributed to a wider investor base after initial underwriting, are instead trading at 89 to 91 cents on the dollar, a signal that the market sees meaningful risk in Oracle's ability to generate returns from its OpenAI tenancy.

The stalled debt sale is not simply a problem with the New Mexico project itself. Rather, it reflects broader anxiety about Oracle's balance sheet as the company has taken on unprecedented debt to build data centers for OpenAI. Oracle's own quarterly filing shows that its $125 billion in senior notes and long-term borrowings carried a fair value of only $105.7 billion as of August 31, 2026, or roughly 84.6 cents on the dollar. The gap between the fixed-rate bonds trading at 85 cents and the floating-rate construction loans trading at 90 cents tells a revealing story: the market is pricing in credit risk, not just interest-rate risk.

What Is Project Jupiter and Why Does It Matter?

Project Jupiter is a 1,400-acre data center campus in Doña Ana County, New Mexico, that Oracle is building specifically for OpenAI as part of the broader Stargate initiative, a joint venture involving OpenAI, Oracle, and SoftBank to develop AI infrastructure. The total project is estimated at $165 billion, with the initial $18 billion construction loan intended to get the campus operational. The site is designed to house massive computing capacity to train and run OpenAI's large language models, the AI systems that power tools like ChatGPT.

The problem is straightforward: the loan is only as good as the tenant paying the rent, and that tenant is OpenAI, a private company that is not yet profitable. S&P, the credit rating agency, flagged this exact vulnerability when it downgraded Oracle to BBB-, one notch above junk status, in July 2026. The agency estimated that roughly half of Oracle's $638 billion backlog stems from the OpenAI contract alone, meaning Oracle's financial health is now deeply tied to a single customer's success.

Why Are Banks Struggling to Sell the Debt?

According to Reuters reporting on Financial Times sources, syndicate banks including Santander and Jefferies are stuck holding more Oracle-linked project debt than planned because efforts to place it with institutional investors have stalled. The banks' silence on the matter is telling; none of the three parties has disputed the 89 to 91 cent quote, which suggests the pricing is accurate and the problem is real.

The core issue is Oracle's rising borrowing and weakening creditworthiness. As Oracle has poured capital into data center construction, its interest expense has surged 55 percent in a single quarter, from $923 million to $1.43 billion. The company's free cash flow turned negative at $5.4 billion in the most recent quarter, even as capital expenditure reached $28.5 billion. Investors are asking a simple question: if Oracle is burning cash at this rate, how confident should we be that it will service this debt?

How Is Oracle Adapting Its Funding Strategy?

Rather than continue down the traditional debt path, Oracle has shifted toward alternative funding mechanisms that move the financial burden away from the company itself. The strategy involves three main levers:

  • Customer Prepayments: Oracle booked $11.4 billion of customer prepayments with a significant financing component in the most recent quarter, compared to none a year earlier. These contracts are discounted at rates consistent with Oracle's own borrowing costs, meaning customers are effectively lending to Oracle through their cloud service agreements.
  • Supplier Financing: Oracle is structuring deals so that suppliers finance the capacity, allowing Oracle to pay for infrastructure as customers pay for it rather than upfront. This spreads the cash outlay over time and reduces immediate capital needs.
  • Customer-Owned Hardware: Some customers are buying the hardware themselves and bringing it to Oracle's data centers, eliminating Oracle's need to finance that equipment.

Oracle also fully deployed its $20 billion at-the-market equity program during the quarter, issuing about 141 million shares for net proceeds of $19.9 billion. This increased shares outstanding by 5 percent in just three months, diluting existing shareholders but raising cash without additional debt. Executive chair Larry Ellison also cancelled his 10b5-1 stock sale plan on September 12 without selling a single share under it, a move that signals confidence in the company's direction even as the stock has faced pressure.

"We said it does not require additional cash from Oracle," stated Co-CEO Clay Magouyrk, explaining that the vast majority of new contracts use prepay, bring-your-own-hardware, or similar mechanics to shift the financial burden away from the company.

Clay Magouyrk, Co-CEO at Oracle

What Are the Physical and Regulatory Challenges at the New Mexico Site?

Beyond the financial hurdles, Project Jupiter faces real-world construction obstacles that could delay completion or increase costs. The New Mexico state land office blocked Oracle's request to run a natural gas pipeline to the campus, forcing the company to switch from 2.2 gigawatts of gas turbines to 2.45 gigawatts of Bloom Energy fuel cells. This substitution adds complexity and cost to the project.

More significantly, the New Mexico Supreme Court paused the air-quality permitting process in August 2026 while it oversees two legal challenges from environmental groups. An Oracle executive told county commissioners that the project was 26 percent complete as of early September, suggesting the timeline remains on track despite the legal challenges, but any major permitting delays could push back the operational date.

Oracle insists the timetable holds. On the fiscal Q1 earnings call, Chief Financial Officer Hilary Maxson stated that the data center is "definitely on track," according to reporting from Ars Technica. However, the combination of stalled debt sales, regulatory uncertainty, and negative free cash flow creates a scenario where Oracle may need to rely even more heavily on customer prepayments and supplier financing to complete the build.

Oracle

What Does This Mean for the Broader AI Infrastructure Buildout?

The Jupiter debt situation is a window into the financial strain that massive AI infrastructure projects are placing on their builders. Oracle is not alone in facing this pressure. The broader Stargate initiative, which aims to build hundreds of billions of dollars in AI computing capacity globally, depends on companies like Oracle, SoftBank, and OpenAI being able to finance construction and operations over many years.

The UAE is also rethinking its approach to AI data center development. After Iranian missile and drone attacks in March 2026 damaged Amazon Web Services data centers in the region, the UAE revised its plans for a 5-gigawatt AI data center project originally designed as a single 10-square-mile campus. The project will now likely be developed as a network of data centers spread across different locations in the UAE, with additional security measures including air defense systems and underground facilities. This shift reflects how geopolitical risk and physical security concerns are now shaping the economics of AI infrastructure investment.

For Oracle specifically, the market's reluctance to buy the Jupiter loans at face value is a clear signal that investors are pricing in real risk. The company's answer has been to shift the burden onto customers, suppliers, and existing shareholders. That strategy may work in the short term, but it also means that if OpenAI's growth slows or if the company faces financial pressure, Oracle's ability to service its debt could come into question far more quickly than traditional credit analysis would suggest.