JPMorgan Chase and Mitsubishi UFJ Financial Group arranged approximately $38 billion in construction financing for two data center campuses linked to Oracle’s expansion plans. Spreading that debt across the market took months.
A Record Deal the Market Had to Digest
Distribution of the Oracle-linked package revealed how selective debt markets have become at this scale and speed.
Loan syndication is standard practice. A bank arranges financing, then sells pieces to other lenders and investors to reduce its own exposure and free capital for the next deal.
The Oracle-linked package covered approximately $23.25 billion for a Texas facility and $14.75 billion in Wisconsin, both senior secured credit facilities tied to Vantage Data Centers. Bloomberg reported that distribution stretched across more than two dozen banks and investors, took months to complete, and left some lenders still seeking to place less than $1 billion well into 2026.
Bloomberg reported pricing at roughly 2.5 percentage points above the relevant U.S. benchmark. That spread reflects construction risk as much as it reflects the effort required to attract enough buyers, though the precise weighting is analytical rather than documented.
Construction exposure runs deeper than assessing Oracle’s creditworthiness alone. Lenders may also face completion risk, permitting timelines, power availability, equipment costs, and the assumption that cash flows will materialize once the facilities are operational.
When Concentration Becomes a Problem
Multiple major banks are reportedly managing the same fundamental tension: they helped build the AI financing machine, and now they need to manage what it produced.
Bloomberg and other outlets have reported that JPMorgan, MUFG, Morgan Stanley, and SMBC explored transactions designed to transfer portions of AI-related loan risk to outside investors while retaining the underlying client relationships. The reported rationale is that internal limits on concentrated AI exposure are becoming relevant.
Concentration risk arises when several large lenders build significant exposure to the same narrow group of borrowers: hyperscalers, data center developers, and AI tenants. If construction costs rise, power access is delayed, or AI demand grows more slowly than projected, correlated losses across multiple institutions become a concern.
Reporting indicates that private-credit funds and non-bank investors are being sought as buyers for debt that traditional bank balance sheets are not retaining. Non-bank buyers typically demand higher yields, tighter covenants, or structural protections in exchange for that residual exposure, shifting pricing leverage away from the original underwriters.
- Banks: risk transfers can free balance-sheet capacity and may indicate that concentration limits are becoming relevant.
- Private credit: increasingly sought as a buyer for debt banks are not retaining, with pricing leverage to match.
- AI developers: more expensive debt could delay construction or require additional equity contributions.
- Data-center developers: projects with secured power and investment-grade tenants are likely better positioned than speculative builds.
Lenders ultimately distributed most of the $38 billion package. Calling this a failure misreads the evidence. The more accurate description is sustained friction: a market working harder than expected to absorb a single financing at unprecedented scale.
The next constraint on the AI buildout may not arrive as a chip shortage or a model limitation. It may appear as a construction loan that takes months to distribute, or a financing cost that makes the math on a new campus harder to justify. Debt markets remain open; they are simply becoming more deliberate about what they will accept, and at what price.
