Banks are actively looking for ways to reduce their exposure to the rapidly growing data center debt market, as the unprecedented scale of borrowing for AI infrastructure development is pushing against their financing limits. Major lenders like JPMorgan Chase, Morgan Stanley, and SMBC are exploring methods to distribute portions of data center-related deals to a wider range of investors.
These efforts include private deals to sell stakes in debt and using risk transfers to free up lending capacity and reduce exposure to large borrowers. The sheer size of these loans is substantial; for example, JPMorgan and MUFG have spent over six months distributing $38 billion in construction debt for an Oracle-leased data center project in Texas and Wisconsin. Some banks even considered selling these Oracle-linked loans at a discount to non-bank lenders.
Lenders are exploring structures similar to significant risk transfers (SRTs), which are typically used by European banks to reduce capital requirements by shifting loan portfolio risks to investors. Unlike traditional SRTs that cover numerous loans, banks are investigating ways to specifically slice and dice large, concentrated data center loans to offload the riskiest parts, while often retaining a percentage of the exposure to assure investors. This strategy is driven by banks hitting risk limits for individual borrowers or sectors, necessitating the freeing up of balance sheets for further lending.
Analysts note that the data center sector, particularly with its construction risk and limited operators, makes these transactions more complex and demands higher compensation for investors compared to traditional SRTs. The financial strain on banks is evident, with one co-head of credit risk sharing at Man Group stating that the scale of borrowing is "out of scale to anything we've thought about, ever," and banks "very quickly start choking." This pressure is leading to a necessity for banks to find more counterparties to finance the market and its pipeline.
Hyperscalers, such as Oracle and CoreWeave, have borrowed hundreds of billions to construct AI data centers, and the aggregate capital expenditure for five major hyperscalers is projected to exceed $690 billion in fiscal year 2026. This surge in AI-driven capital expenditure, coupled with declining free cash flows, is compelling these companies to increasingly rely on external financing through debt issuance, equity raises, and large-scale leasing arrangements to fund their AI infrastructure. Total debt for these hyperscalers has reached approximately $700 billion, with a significant portion of their lease obligations (around $1.1 trillion of $1.4 trillion disclosed) currently off-balance sheet.
