AI Bonds Are Starving the Treasury of Capital

The Treasury Is Losing a Bidding War It Was Never Meant to Lose

The United States now carries $40 trillion in debt. Federal borrowing alone is on pace to hit $2 trillion this fiscal year, and interest payments on that mountain are already consuming $1 trillion annually. In any normal cycle, that kind of sovereign demand would squeeze every other borrower out of the market. Companies would pay more to borrow. Yield spreads would blow out. Risk would reprice across the board.

That isn’t happening. Not because the federal government isn’t competing for capital, but because a different kind of borrower has stepped into the ring — one so hungry for chips, data centers, and GPU clusters that it is willing to absorb debt at yields that would have spooked investors two years ago.

AI hyperscalers are issuing corporate bonds at a clip that is distorting the entire fixed-income landscape.

How the Mechanism Works

U.S. investment-grade corporate bond issuance has reached roughly $1.7 trillion through July, up 27 percent from last year and tracking toward a first-ever annual figure above $2 trillion, according to Wall Street veteran Ed Yardeni.

At that volume, you would expect corporate yields to rise relative to Treasuries — investors demanding a bigger premium for absorbing that much debt. Instead, the spread has stayed stubbornly compressed. Demand for AI-related bonds has been fierce enough that hyperscalers aren’t paying dramatically more than they did before the spending surge began.

So who is paying the price? The Treasury.

Capital that would traditionally flow into sovereign debt is being pulled toward corporate issuance. Treasury yields have had to climb to attract buyers, not because the economy is overheating alone, but because the bond market is clearing a new equilibrium — one where private AI infrastructure is effectively competing with the federal government for finite pools of capital.

This is the reverse-crowding-out effect, and it is unlike anything the bond market has experienced.

Why Investors Are Buying

The appetite comes down to one word: returns.

AI hyperscalers — the companies building the data-center megaprojects that underpin the generative-AI boom — believe the economics of their infrastructure spend are so favorable that borrowing costs are almost irrelevant. They need capacity now. The window for securing land, power, and GPUs is narrow. Delay means falling behind competitors who are already online.

Treasury Secretary Scott Bessent acknowledged the dynamic directly, noting that much of the corporate issuance coming out of AI companies is yield-agnostic. They aren’t shopping for cheap money. They are shopping for available money.

That changes the mathematics of the market. When the primary buyers don’t care about yield, the marginal buyer does — and that buyer is increasingly forced into Treasuries, which then require higher yields to move.

Foreign Money Has Already Picked a Side

The clearest signal of this shift isn’t domestic. It is international.

Data from capital-flow trackers show that over the past year, foreign private-sector buyers have purchased more U.S. corporate bonds than U.S. Treasuries. That is a striking reversal. For decades, foreign official and institutional investors have been the backbone demand side of the Treasury market — central banks, sovereign wealth funds, pension funds parking dollars in the safest instrument available. Now they are rotating toward corporate credit, and specifically toward corporate credit tied to AI.

This isn’t a trivial rotation. It means the structural bid for Treasuries that has held yields down for years is weakening at the exact moment the federal government needs it most.

The Feedback Loop Nobody Is Pricing In

Here is where the story gets uncomfortable for policymakers.

Higher Treasury yields increase the government’s debt-servicing costs. More spending on interest widens the deficit. A wider deficit requires more issuance. More issuance pushes yields even higher. The loop feeds itself.

Yes, other forces are at play — oil prices from the Iran conflict, persistent inflation, a robust economy all putting upward pressure on rates. But the AI corporate-debt surge is an independent variable that makes every one of those problems harder to manage. It is not a cyclical bump. It is a structural reallocation of capital away from sovereign debt and toward private infrastructure buildout.

The Treasury Secretary has noticed. Fed Chairman Kevin Warsh flagged the same dynamic at Jackson Hole this past Friday, describing ever-expanding pools of capital pouring into AI infrastructure. Both men understand the direction of travel. Neither has offered a way to slow it down — and that, in itself, is telling.

Where the Market Is Showing Fray Lines

Don’t mistake compression for invincibility. S&P Global warned last month that the market is beginning to show signs of fatigue after absorbing the flood of debt in such a compressed window. Hyperscalers are already paying a wider premium compared to risk-free yields than they were six months ago. The spread hasn’t blown out, but it has moved. It is moving.

S&P noted that investors are growing leery of rapidly rising leverage from issuers previously characterized by strong, reliable cash flow. That description applies to the very companies driving this entire phenomenon. If the narrative shifts from AI infrastructure as a growth engine to AI infrastructure as a balance-sheet burden, the repricing could be swift.

And it wouldn’t be contained to corporate credits. A disorderly adjustment in AI corporate bonds would ripple directly into Treasuries — potentially lowering yields there in the short term, but at the cost of a broader credit crunch that the Federal Reserve would feel immediately.

Hidden Borrowing Compounds the Risk

The bond-market data captures only the visible half of the picture. Private credit financing for AI is exploding. Nvidia is leveraging its balance sheet to back AI deals directly. And so-called hidden borrowing — structures, vehicles, and contingent obligations that don’t show up cleanly on traditional balance sheets — has been tallied at approximately $1.65 trillion.

That figure is not an error. It is a reflection of how the AI buildout is being funded: through every channel available, and through channels that many investors don’t fully understand. Hidden borrowing doesn’t disappear when rates rise. It simply migrates further out of visibility until the visibility gap becomes a solvency gap.

What This Means for Global Investors

The reverse-crowding-out dynamic has implications that extend well beyond Washington.

If corporate bonds are absorbing capital that once flowed into Treasuries, then the global safe-asset supply is shrinking at the same moment demand for it should be rising — given the geopolitical turbulence (Iran war, oil shocks) and domestic fiscal trajectory already in place. That creates a bid for Treasury yields that is structural, not seasonal.

For fixed-income managers, the immediate play is straightforward: Treasuries are no longer the passive recipient of global capital. They are in active competition with some of the most aggressively funded companies on Earth, and those companies are winning that competition right now.

For policymakers, the problem is sharper. Every point of yield on the 10-year Treasury represents additional annual interest cost the federal government must fund. At current levels, that math is already strained. At higher levels, it becomes self-reinforcing.

The Unpriced Risk

What Western investors seem to be missing — or at least what hasn’t been fully incorporated into pricing — is that this is not a temporary spike in corporate issuance. It is a sustained, multi-year reallocation of debt capital toward AI infrastructure. Data-center construction timelines run for years. Chip orders are placed on multi-year horizons. The borrowing doesn’t stop when the initial announcement cycle ends.

The market has absorbed the first wave. The second wave is coming. And by the time investors price in the cumulative impact of both waves on Treasury demand, the yield damage may already be locked in.

The reverse-crowding-out effect is real. It is structural. And it is reshaping the U.S. debt market from the ground up — with the federal government on the losing side of a bidding war it was never designed to fight.