How Much Debt Have AI Hyperscalers Issued?

Robin Wigglesworth says AI hyperscalers have issued around $500 billion worth of debt in the latest year, after about $200 billion the year before. He is not actually worried, because the borrowers are large, successful companies. YouTube still prints money. Amazon still prints money. Facebook and WhatsApp still print money. They could issue lots more.

A lot of money is going into AI, and the spending is increasingly debt financed and off-balance sheet financed. Historically, that pattern has not ended well. Rail was a capital-spending boom that ended badly in economic and financial terms and still brought a huge societal benefit for America, and so did all the cable that was laid. Rail and cable last longer than a GPU chip, which is the part that worries him.

Rating agencies are looking into the off-balance sheet commitments, and the liabilities are getting integrated. The companies remain pretty strong. Oracle looks like the weaker member of the herd, and even Oracle does not look like an imminent disaster. The buildout is still in the foothills. He struggles to see the borrowing ending horrifically badly. He has been wrong before, and a year on the verdict might be that it broke pretty bad.