Tighter data center credit rewards the biggest clouds

The banks that competed to fund AI data centers are now trying to sell parts of those loans. According to the Financial Times, as reported by Startup Fortune, JPMorgan, Morgan Stanley, SMBC and MUFG have explored significant risk transfers. In these deals the bank keeps the loan but moves its riskiest slice to outside investors. The stated reason is that the banks are approaching internal limits on AI exposure. Bloomberg reports that one lender group has spent more than six months trying to place $38 billion of construction debt for Oracle-leased sites in Texas and Wisconsin, with some lenders willing to sell at a discount.

Bloomberg also reports capital rotating toward data centers leased to Microsoft and Google. A gap of a point or two on the borrowing rate compounds over a multi-year build. If it holds, smaller GPU clouds either pass a higher cost of capital into their prices or scale back. Compute then keeps concentrating in the companies that already own most of it, for reasons that have nothing to do with whose stack works better.

If you are signing a multi-year deal for computing capacity, how the provider pays for its data centers now deserves the same scrutiny as its uptime guarantees. Some of the warning signs are public. In early October, the cost of insuring Oracle's debt against default over five years rose to about 1.5% of the amount covered per year, a level last seen in 2009, according to Bloomberg. S&P rates Oracle BBB-, the lowest grade still considered investment grade. The practical questions are what happens to the capacity you reserved if the provider has to borrow again while interest rates are high, and how quickly you could move your workloads if it does.