AI's $1.7 Trillion Bond Wave Is Now Lifting Everyone's Cost of Borrowing

The 30-year Treasury touched 5.323% on August 18, the highest yield on a U.S. government bond of that maturity since 2007. Two days later, the 10-year sat at 4.69%. By that point, U.S. corporations had already sold nearly $1.7 trillion of bonds in 2026, up 27% from a year earlier and more than all of 2025 combined, according to data from the Securities Industry and Financial Markets Association cited by 247 Wall St. The two stories are connected, and the connection is not a comfortable one for the same companies doing the borrowing.

AI capex is being paid for with debt. The spending covers graphics processors, the data centers that house them, the power plants and substations that feed them, and the long-term cloud contracts that lock in capacity. That money is not sitting in a corporate checking account. It is being raised in the bond market, much of it in maturities of ten years or longer. That is the same corner of the market where pension funds, insurance companies, and foreign central banks park the bulk of their safe, long-dated assets, so a flood of new corporate supply there has nowhere to hide.

BMO Capital Markets rates strategist Ian Lyngen frames the mechanism as a supply problem at the long end of the Treasury curve. When a heavy calendar of long-dated corporate bonds lands in a market already digesting record U.S. Treasury issuance, investors demand more interest to absorb it. The extra yield they demand for locking up money for decades is the term premium, and the long end of the curve is where it shows up first. Higher long-end yields then feed into mortgages, car loans, and the corporate borrowing costs that price almost every other deal in the economy.

The chip stocks and the bond market are not separate stories. NVIDIA closed at $216.85 on August 20, with a market cap above $5.26 trillion. Its first quarter of fiscal 2027 brought in $81.61 billion of revenue, up 85% year over year, with $75.25 billion of that from data center chips. The company itself carries almost no debt relative to equity and earns hundreds of times its interest expense in operating profit, so it is financing its own capex internally. Its customers are not. The deals NVIDIA has signed with Meta, Anthropic, CoreWeave, AWS, Google, Microsoft Azure, and Oracle imply hundreds of billions of dollars of GPU purchases, and a meaningful share of that tab is being financed by the same customers selling bonds into the market the AI buildout itself is stressing.

Digital Realty Trust, the real estate trust that owns many of the actual buildings holding the servers, is the cleanest example of an issuer being forced into the market by the AI buildout. Higher long-end yields raise the cost of the long-term debt that funds a new data center shell, while the same higher yields make the trust's dividend-heavy equity less attractive to income investors. Squeezed on both ends, the trust keeps issuing debt to keep building.

The loop is the part investors should hold onto. Higher long-end yields raise the discount rate applied to the future earnings that justify today's AI valuations. A discount rate that rises because AI capex is borrowing at the long end of the Treasury curve is, in effect, the AI cohort pricing its own multiple compression. Jensen Huang has called the current buildout "the largest infrastructure expansion in human history." The bond market is asking investors to decide how much of that history they want to fund at 5.3% for thirty years.

The next test of the loop arrives whenever the long end jumps again. The share of the move that is AI-linked corporate issuance, the share that is U.S. fiscal supply, and the share that is oil or inflation expectations will not show up in the headline number. They will show up in who pays and who benefits, and that mix is now the most useful thing a reader can track through the rest of the year.