Big Tech Doubles Debt to $350B to Fund AI Data Center Buildout

For years the hyperscalers' story ran on a reassuring line, that they were so cash-rich they could fund any buildout out of pocket. That line is now under visible pressure. Bloomberg reported on Friday that Alphabet, Amazon, Meta, Microsoft and Oracle have collectively added roughly $350 billion of debt over the last five years to keep pace with AI capital spending, and that the five together now pay more than $10 billion a year in interest, more than double 2019 levels.

The pace of spending is the thing that broke the old model. The five have pledged up to $725 billion this year alone on data centers and Nvidia chips, according to Bloomberg reporters Matt Day and Ian King. Google's free cash flow was $64 billion at the end of the March quarter, and Amazon's went negative in that same quarter, which is the sort of number that changes how debt investors price the paper. Amazon's recent $25 billion bond offering got what the reporters described as an unusually chilly reception. Oracle, which now carries debt of roughly 2.5 times its 2025 sales, was cut by S&P to the lowest investment-grade rating on Thursday.

Why this matters if you don't hold the bonds: the AI buildout has been sold as a self-funding transformation of the economy, and it is now visibly being funded by lenders. Jason Pompeii of Fitch Ratings told Bloomberg, "I don't know that we know whether Amazon, Google, Microsoft and Meta are actually going to get a return on investment." Gil Luria of DA Davidson described the shift more bluntly, saying, "The nature of these businesses is changing very dramatically, and it's changing abruptly. That's why their cash flow is so depressed right now." The CEOs are still calling the top of the demand curve. Amazon's Andy Jassy told shareholders in April he has "high confidence this will be monetized," and Meta's Mark Zuckerberg has said AI compute demand continues to exceed supply.

The honest caveat is that this is a snapshot of a fast-moving credit picture, and the reporting is thin on the counterfactual, what these balance sheets look like if AI revenue does ramp on the schedule the CEOs are guiding to. What the piece doesn't give you is a clean line between capex committed and revenue realized at each of the five, or how much of the debt is short-dated versus locked in long. The interesting thing to watch from here is whether Oracle's downgrade sits alone or is the first of a set, because that is the point at which the story stops being about ambitious spending and starts being about the cost of capital.