AI companies are borrowing like the U.S. government. That's now a problem for the U.S. government.

U.S. investment-grade corporate bond issuance is on pace to clear $2 trillion for the first time, according to a Fortune tally of data attributed to Ed Yardeni, running roughly 27% ahead of last year's pace through July. A large share of that wave is coming from a small set of AI hyperscalers. Those are the cloud and model labs spending hundreds of billions on chips, data centers, and the power contracts behind them. Treasury Secretary Scott Bessent now says, on the record, that those issuers are "almost yield-agnostic": they don't seem to care what interest rate they pay.

When private borrowers stop pricing the cost of money, the market that clears the U.S. government's own debt has to adjust. That is the mechanism Bessent's framing points to, and it is the one Wall Street veteran Ed Yardeni has been calling, in a phrase, a "reverse crowding-out" effect. The classic economics-textbook version runs the other way: heavy government borrowing pushes interest rates up and squeezes private investment. Here the direction flips. Private AI borrowing is pulling capital toward corporate bonds and away from Treasuries, so Treasury yields have had to rise to clear the market. Yardeni's label is interpretive, not a consensus term, but the arithmetic is what it is.

"Yield-agnostic" means the AI issuers are issuing debt at coupons they would not have accepted for an industrial project with normal expected returns. They believe the build-out will pay back fast enough that the coupon is a rounding error. Bessent put it this way: "We are also seeing big corporate issuance. And a lot of that corporate issuance, I would say, is almost yield-agnostic, because the build-out for AI, the returns on that, the companies believe they're going to be so high. They don't really care what they're paying."

U.S. federal debt stands at $40 trillion, annual debt-servicing costs are running about $1 trillion, and the budget deficit is on track to reach $2 trillion this fiscal year, per the same Fortune reporting. Against that backdrop, an investment-grade corporate calendar closing in on $2 trillion, a record, is not a footnote. Fidelity's Jurrien Timmer posted the corporate-bond data point as a piece of the broader market picture. The demand has been strong enough that yield spreads on AI-related issuance over Treasuries have stayed compressed, according to an Investing.com analysis, rather than widening to absorb the supply. That is the behavior a "yield-agnostic" buyer would produce: tight spreads even as the calendar swells.

The risk Bessent's framing surfaces is a feedback loop, and it is forward-looking rather than measured. Higher Treasury yields raise the government's debt-servicing bill, which widens the deficit, which puts more Treasury supply into the market, which puts upward pressure on yields again. Edward Conard's macro note argues the AI capex cycle is on a scale that is already reshaping who buys what, and Bessent's Jackson Hole-adjacent messaging has been read by commentary on the Fed path as a signal that AI capex is now part of the rate discussion. None of that proves the loop has closed. It does mean the auction calendar, the term premium, and the deficit are now all moving in the same direction the AI channel wants them to move.

Fortune's own piece names other drivers of higher Treasury yields: the deficit itself, the term premium, and Fed policy. If the AI-bond channel is a co-driver rather than the primary one, "reverse crowding-out" is a partial read, not the whole story. Yardeni's frame is a labeled interpretation; Bessent's "yield-agnostic" is a characterization of issuer behavior, not a measurement of how much of the yield move is attributable to AI demand. The story is that the channel exists and is large enough for a Treasury Secretary to name, not that it is the only channel.

What to watch, then, is narrow. The IG corporate calendar's pace through the back half of the year. The spread between AI-heavy IG new issues and on-the-run Treasuries. Treasury auction tails and the term premium on the long end. And Bessent's next public framing: if he moves from "yield-agnostic" to a specific number, the story has graduated from a labeled mechanism to a measured one.