Major financial institutions, including JPMorgan, Morgan Stanley, SMBC, and MUFG, are currently attempting to distribute large portions of AI-related construction debt that they previously underwrote. While underwriting—the process of guaranteeing a loan—is a standard banking function, the current difficulty in 'offloading' or selling these loans to outside investors suggests a tightening of risk tolerance. According to reports, a group of these banks has spent over six months attempting to distribute $38 billion in debt tied to a data center project in Texas and Wisconsin leased to Oracle. This delay is notable because such syndications typically clear much faster, indicating that the market for this specific type of high-scale AI debt is becoming more constrained.
This image illustrates the broader banking landscape and the infrastructure projects currently facing financing hurdles.
The primary mechanism driving this shift is the exhaustion of internal risk limits. Banks often face regulatory or internal caps on how much exposure they can hold in a single sector. To manage this, they use risk transfers to move the riskiest portions of a loan to third-party investors while keeping the loan on their books. However, the sheer scale of the AI buildout—estimated at over $120 billion in infrastructure spending—has pushed some lenders toward their limits. This is compounded by a cooling demand for AI-related bonds; while these securities previously sold themselves, order books have shrunk significantly, forcing banks to work harder to place the debt. Furthermore, the rising cost of capital, evidenced by high 10-year Treasury yields, makes the 'risk-free' rate a higher hurdle for financing these massive projects.
Market participants are also becoming more selective about which AI projects they fund. Capital is increasingly rotating toward 'hyperscalers'—large, investment-grade tech giants like Microsoft and Google—and away from smaller, less capitalized infrastructure companies. For example, debt for companies like CoreWeave, which lease to smaller entities, is commanding significantly higher interest rates than bonds leased to major hyperscalers. This creates a bifurcated market where smaller AI infrastructure players must compete for a shrinking pool of lenders without the balance sheet strength of a company like Oracle. The struggle to move $38 billion in debt over six months serves as a practical indicator that the assumption of indefinite, low-cost credit for AI infrastructure is facing real-world friction.