Oracle-linked $18bn Project Jupiter debt slips to 89-91c as New Mexico risks mount

Oracle Corporation (NYSE: ORCL) is facing a new financing test after approximately $18 billion of loans tied to an Oracle-leased artificial intelligence data-centre campus in New Mexico were quoted by arranging banks at 89 to 91 cents on the dollar. Santander and Jefferies Financial Group Inc. were among the institutions marketing the debt, but efforts to distribute it to a broader investor base stalled amid concern over Oracle’s rising borrowing, weakening credit profile and local opposition to the project, the Financial Times reported. The 1,400-acre Project Jupiter campus in Doña Ana County is intended to provide computing capacity under Oracle’s wider relationship with OpenAI. Oracle, Santander and Jefferies did not comment to Reuters, which summarised the Financial Times report.

The discount does not mean that Oracle has defaulted or that lenders expect an immediate failure. It means the loans are being valued below their face amount as buyers demand compensation for perceived credit, construction, permitting and liquidity risk. Banks that intended to syndicate a larger portion may have to keep more exposure on their own balance sheets, sell at a loss or wait for conditions to improve. The episode shows how the artificial intelligence infrastructure boom is moving risk from technology-company forecasts into project finance, utility planning, local politics and bank capital.

What does a price of 89 to 91 cents mean for the $18 billion loan package?

A loan quoted at 89 cents is worth $89 in the secondary or syndication market for every $100 of face value, before transaction details and accrued interest. The discount indicates that prospective investors want a higher yield or a lower entry price than the original terms provide. It can reflect concern about the borrower, the tenant, construction delays, collateral, contract structure or the ability to resell the debt. It is a market signal of stress, but not by itself evidence that scheduled interest or principal has been missed.

For arranging banks, the difference can create an underwriting problem. Institutions often commit financing with the expectation that they will distribute much of it to loan funds, insurers and other investors. If demand weakens, they may remain exposed longer than planned and must allocate balance-sheet capacity to the position. Selling at 89 to 91 cents crystallises a discount, while holding the loans leaves the banks sensitive to further deterioration or a recovery in price.

The legal relationship between Oracle and the loan borrowers is also important. Reuters described the financing as loans tied to an Oracle-leased data centre rather than ordinary unsecured Oracle corporate bonds. Project vehicles, developers, lease obligations, guarantees and customer contracts can distribute risk differently. Investors should not automatically add the full $18 billion to Oracle’s reported corporate borrowings or assume that Oracle is legally responsible for every dollar in the same way as a senior note. The credit concern arises because Oracle’s lease and computing commitments are central to the project’s economics even if the financing sits in a separate structure.

Why has local opposition become a financing risk for Project Jupiter?

The campus was initially expected to use 2.2 gigawatts of gas-turbine generation, an extraordinary power requirement comparable with the output of multiple large power plants. New Mexico’s state land office blocked a request for a natural-gas pipeline serving the site, according to the Financial Times report, complicating the proposed energy plan. Residents and campaigners have raised concerns about water supply and air quality, while legal and political opposition has increased uncertainty around permits and timing. A gubernatorial candidate has called for a moratorium on new data centres, adding a state-level policy risk to local objections.

Every permitting delay can affect credit because a data centre cannot generate contracted capacity until land, power, cooling, network and computing equipment are ready. Construction costs and interest can continue while revenue moves further into the future. If the energy design must change from gas turbines to another source, developers may face new equipment, grid-interconnection and permitting requirements. A lender that expected a defined completion schedule must then reassess contingency, cash interest and the strength of completion guarantees.

Community opposition is not merely a public-relations issue when resource demand is this large. Water availability can constrain cooling and construction, while air permits determine whether on-site generation can operate. Local governments may support jobs and tax revenue but still impose conditions on emissions, noise, infrastructure or consumption. The gap between a hyperscaler’s desired build speed and a community’s approval process can become a direct financial variable, which is why Project Jupiter’s local debate is now visible in the loan price.

How much financial capacity does Oracle have for its artificial intelligence build-out?

Oracle’s latest results show both powerful demand and intense capital pressure. In the first quarter of fiscal 2027, revenue rose 30% to $19.3 billion, total cloud revenue increased 62% to $11.6 billion and cloud infrastructure revenue climbed 121% to $7.4 billion. Remaining performance obligations reached $664 billion, up $209 billion from a year earlier, after Oracle booked more than $30 billion of additional artificial intelligence cloud contracts. The company also said it delivered 850 megawatts of extra data-centre capacity during the quarter.

Those growth rates require enormous upfront investment. Oracle reported a record $23 billion of operating cash flow in the quarter but negative free cash flow of $5 billion after capital expenditure. It sold $20 billion of common stock through an at-the-market programme as part of its disclosed capital plan. Non-current notes and other borrowings stood at approximately $117.7 billion at the end of August, while quarterly interest expense rose 55% to $1.43 billion. These figures from Oracle’s first-quarter fiscal 2027 results explain why creditors are examining execution and funding as closely as the revenue backlog.

Demand and credit quality can move in opposite directions. A vast backlog supports the argument that data centres will have customers, but building enough capacity to serve those contracts can require debt, equity, leases and partner finance before revenue arrives. Oracle’s ability to convert remaining performance obligations into cash depends on construction schedules, equipment delivery and customer use. The Project Jupiter loan discount signals that some investors are no longer treating contracted artificial intelligence demand as sufficient protection against every site-level and balance-sheet risk.

How does Project Jupiter connect Oracle with OpenAI?

Project Jupiter is part of Oracle’s wider agreement to provide artificial intelligence computing capacity to OpenAI. The economic logic rests on matching a long-duration customer demand commitment with a purpose-built campus, specialised chips and large power supply. Oracle gains cloud revenue and a stronger position against Amazon Web Services, Microsoft Azure and Google Cloud, while OpenAI secures capacity for model training and inference. The relationship is commercially powerful but concentrates execution risk around a small number of exceptionally large customers and projects.

OpenAI has said its broader Stargate effort surpassed an initial 10-gigawatt United States infrastructure commitment ahead of schedule and is evaluating additional sites. That ambition supports demand for partners across data centres, energy, chips and finance. It also intensifies scrutiny over who bears construction, utilisation and counterparty risk. A contract can be valuable while still leaving lenders exposed if a specific campus is delayed or if its power solution changes.

The $18 billion financing therefore matters beyond New Mexico. If banks cannot distribute a loan package backed by one of the most visible artificial intelligence infrastructure relationships, future projects may require higher yields, more equity, stronger guarantees or tighter covenants. Those changes would raise the cost of capacity and could slow deployment. A successful restructuring or completion plan, by contrast, could show that temporary syndication stress can be resolved once permits and energy arrangements become clearer.

What does Oracle’s credit rating add to the concern?

S&P Global Ratings downgraded Oracle in July, leaving the company’s corporate rating one notch above speculative grade, according to the Reuters report. A rating near the investment-grade boundary makes creditors sensitive to additional leverage, negative free cash flow and execution setbacks. If perceived risk increases, the cost of corporate bonds, project loans and lease financing can rise even when operating revenue is growing rapidly. Some regulated or mandate-constrained investors may also reduce exposure if debt falls below investment grade.

Oracle’s equity issuance shows that management is using more than one funding channel. Selling $20 billion of stock reduces the amount that must be financed through debt and can protect credit metrics, but it dilutes existing shareholders. The company has said the structure of new contracts does not create an incremental need beyond its capital-raising plan. Project-level stress will test whether that assurance remains credible as multiple campuses move from announcement to construction.

The key credit question is not whether artificial intelligence demand exists. Oracle’s 121% infrastructure growth and $664 billion backlog provide strong evidence of customer commitments. The question is whether revenue, financing and construction cash flows are timed and structured so that the company and its partners can fund capacity without eroding credit quality. Project Jupiter’s discounted loans suggest that investors want a larger margin for error.

How did Oracle shares respond to the debt-pressure report?

Oracle shares closed at $147.61 on 18 September, down about 2.0% for the session. The Financial Times report was published late in the trading day, so the move may reflect a combination of the project news, interest-rate conditions and existing concern about capital intensity. The stock remained under pressure despite Oracle’s rapid cloud growth because investors have been balancing unprecedented backlog against negative free cash flow, equity issuance and a heavier financing burden. One session cannot isolate the value of Project Jupiter, but the direction is consistent with that debate.

The stock reaction also differs from the loan signal. Equity holders participate in the upside if Oracle converts artificial intelligence contracts into high-growth cloud revenue, while lenders focus first on repayment and downside protection. A project delay can therefore hurt loan pricing even if long-term equity investors remain optimistic about demand. Conversely, a stronger balance sheet can support credit while share issuance weighs on per-share value. Reading both markets gives a more complete view of the financing trade-off.

Jefferies Financial Group Inc. and Santander face their own exposure as syndicate banks, but the report does not quantify each institution’s retained position or potential loss. It would be inaccurate to infer material damage to either bank from the headline loan total alone. Disclosure of final distribution, marks or provisions would be needed to assess the effect. For now, the syndication difficulty is evidence of investor caution rather than a disclosed banking loss.

What milestones could restore confidence in Project Jupiter financing?

The most important milestone is a bankable power plan. Developers need clarity on whether the 2.2-gigawatt gas proposal can be modified, replaced or permitted and how the chosen solution affects cost, emissions and completion timing. Resolution of water, air-quality and land-use issues would reduce the range of possible delays. A public construction schedule with achieved milestones would help lenders distinguish political noise from an actual threat to delivery.

Loan distribution will provide a second signal. If banks place debt near par after permits improve or terms change, the current discount may prove temporary. If prices fall further or lenders retain unusually large positions, concern is likely to spread to other artificial intelligence project financings. Changes in covenants, guarantees or sponsor equity would reveal what protection investors require to re-enter.

Oracle’s corporate metrics form the final test. Positive free cash flow, stable investment-grade ratings and evidence that contracted capacity is entering service would reduce fear that growth is outrunning funding. Continued negative free cash flow, further downgrades or repeated site delays would reinforce it. Project Jupiter is not simply a story about one loan package. It is a live demonstration of whether the capital markets can finance artificial intelligence infrastructure at unprecedented scale when physical-world constraints begin to challenge digital demand forecasts.