The same AI capex is now both the floor under US growth and the bond market's biggest bet

The same AI capital spending that now props up roughly a third of US economic growth is, by Treasury Secretary Bessent's own public account and a Bank of America fund manager survey, the most plausible single trigger of a corporate-bond market panic.

Goldman Sachs Research projects about $1 trillion of AI-related investment globally in 2026, including $581 billion in the US. The bank argues this sits above the more commonly cited ~$800 billion hyperscaler capex consensus because the true AI investment line counts power, land, grid upgrades, and supply-chain build-out, not just the chip and data-center spend that shows up on hyperscaler balance sheets.

ING's James Knightley wrote on 20 August that data centers and power generation are now the only construction sub-sectors still growing. Real computing and software investment is up a cumulative 62% since the first quarter of 2022. Non-tech, non-residential business capex has fallen year over year for six straight quarters. Strip AI out of the construction and capex lines and the 2026 US growth print thins visibly. ING estimates AI may account for roughly a third of US growth in 2026. The wealth-effect channel does more of the work than direct consumer subscription spend: AI's contribution shows up in construction payrolls, grid investment, and the equity holdings of households exposed to AI-adjacent stocks, not in monthly app receipts.

The capex is also doing the work on the equity side. Goldman Sachs's Ben Snider told CNBC, as reported by 247 Wall St on 13 September, that AI capex is driving roughly half of S&P 500 earnings growth, even as the index's price-to-earnings multiple has compressed from about 22 times to about 19 times. The 10-year Treasury yield sits at 4.83%, the 99.6th percentile of the past year. Earnings rising on AI build-out while the multiple contracts is the equity version of the same paradox: the growth is real, the price for it is going up, and the spend shows up in the cashflow statement that the bond market is now pricing.

Bessent spoke into that backdrop. On The Spillover, the Council on Foreign Relations podcast, he made the unusual suggestion that AI companies should stop issuing so many long-term bonds and instead issue medium-term ones. The same episode notes that Treasury yields have risen to levels not seen since 2007. "Duration" is the bond-market term for how much an issuer's payments are exposed to interest-rate moves. A 30-year bond has more duration than a five-year note, and more price risk if yields keep climbing. Bessent's nudge was, in effect, a public warning that the duration of AI-related debt is now a Treasury concern, not only a corporate finance one.

The bond market has heard him. The CFR episode cites a recent Bank of America fund manager survey in which almost 40% of respondents named AI hyperscaler capex as the most likely source of a panic in the debt market, a higher share than any other single candidate. Re-reporting in Moneycontrol tracks the same mechanism without adding new data points.

The mechanism, then, is one trade viewed from two desks. AI capex is the floor under US GDP, the only construction sub-sector still expanding, and roughly half of S&P 500 earnings growth. It is also the duration load being layered onto the corporate bond market at the moment long yields sit at a one-year extreme and the curve has already repriced. The trigger is not a single number. It is a refinancing wall meeting a yield curve that has moved, with the bond market now pricing the meet.

Watch Bessent's next Treasury remarks on corporate-bond duration, the tail on the September 2026 10-year auction, the Q3 hyperscaler earnings calls, and the Q4 corporate-bond issuance calendar. If any of them print against the AI-capex-as-floor view, the floor and the bet move together.