Summary
- Roughly $18bn of loans linked to Project Jupiter are being quoted at 89–91 cents on the dollar.
- The New Mexico campus forms part of Oracle’s broader AI infrastructure relationship with OpenAI.
- Debt distribution has been complicated by Oracle’s borrowing, environmental opposition, and uncertainty around energy infrastructure for the project.
About $18bn of loans tied to an Oracle-leased data centre development in New Mexico are trading below par as banks face growing scrutiny of the financing and infrastructure assumptions behind large AI campuses.
The debt is linked to Project Jupiter, a proposed 1,400-acre campus in Doña Ana County being developed to support Oracle’s wider AI computing arrangements, including capacity associated with OpenAI.
Reuters, citing Financial Times reporting, said banks were quoting the loans at around 89 to 91 cents on the dollar after efforts to distribute the debt more widely stalled.
Santander and Jefferies are among the banks involved in the lending syndicate. The institutions and Oracle declined to comment on the report.
The pressure around the debt is notable because financing risk is increasingly becoming inseparable from power and planning risk in large AI data centre developments.
Project Jupiter has faced local opposition centred on water and air-quality concerns, while proposed energy infrastructure has also become a point of uncertainty. An attempt connected with gas infrastructure intended to serve planned on-site generation has encountered regulatory difficulty.
For lenders, those issues affect more than the eventual opening date. A delayed power solution or permitting process can shift construction schedules, extend the period before a facility generates revenue, and increase the amount of capital tied up in a project.
That exposure becomes more significant when individual campuses are financed with debt measured in tens of billions of dollars.
Oracle has been expanding aggressively to support AI computing demand, increasing both its capital requirements and the amount of infrastructure-associated borrowing that financial markets have to absorb.
The Project Jupiter loans were originally arranged by a banking consortium with the expectation that portions could be distributed to other investors. Difficulty doing so leaves originating banks holding more exposure than planned and makes the market price of the loans a useful indicator of changing risk appetite.
Credit risk is only one part of the equation. Very large AI facilities also depend on customers continuing to require the contracted capacity and on computing hardware retaining enough economic value to support long-duration financing structures.
The rapid turnover of AI accelerators adds another complication. Data centre buildings, substations, pipelines, and transmission infrastructure can have useful lives measured in decades, while the highest-value computing equipment may move through technology generations far more quickly.
This creates a financing structure in which long-lived physical infrastructure is being built around a fast-changing technology market. Contract length, customer credit quality, residual equipment value, and the ability to repurpose facilities all become important to lenders.
The issue has direct relevance to European infrastructure even though Project Jupiter is in the US. European AI campuses are also moving towards larger capital packages, more complicated project-finance structures, and stronger dependence on long-term power and customer commitments.
A project can have an apparently strong demand case and still encounter difficulty if the electricity solution, permits, community acceptance, or debt market moves against it.
Project Jupiter therefore provides an unusually large example of a more general pressure facing the sector: AI demand can support unprecedented data centre investment, but the cost of capital still depends on how convincingly developers can turn an announced campus into powered, permitted, revenue-producing infrastructure.
