The surge in U.S. Treasury yields to levels not seen in two decades has sparked intense debate over its drivers and durability. While inflation has persistently run above the Federal Reserve's 2% target, the recent climb in yields—since late August—has not coincided with rising inflation expectations. Instead, the move is largely attributed to a jump in real yields, reflecting stronger underlying economic fundamentals rather than price pressures.
Fed Chair Kevin Warsh, in his September press conference, pointed to three forces: robust economic growth, a surge in capital spending, and geopolitical tensions. The first two are intimately tied to the unprecedented expansion of artificial intelligence infrastructure. This buildout is now supplementing consumer spending as a key engine of U.S. growth, as I previously noted.
The scale is staggering. The five major hyper-scalers—Amazon, Alphabet, Meta, Microsoft, and Oracle—are projected to spend over $800 billion this year and $1.1 trillion next year on AI-related capital projects. Columbia Business School professor Stijn Van Nieuwerburgh puts this in historical context: the pipeline of planned investments could total more than $10 trillion from 2025 to 2032, equivalent to 3.6% of GDP annually. That would dwarf past infrastructure booms in canals, railroads, electrification, highways, and telecommunications.
However, these GDP-relative figures may overstate the domestic economic impact, as a significant share of spending leaks into imports. So far, the growth effect has been driven primarily by business investment. But as AI adoption spreads and productivity gains materialize, the impact could deepen. A Federal Reserve Board study suggests adoption is still in a buildout phase, with productivity gains concentrated in certain sectors and labor market disruptions not yet broad-based.
The financing demands of this buildout are now colliding with a strained federal fiscal picture. With the budget deficit hovering around 6% of GDP and public debt near record highs, competition for capital is intensifying. Morgan Stanley estimates that hyper-scalers will need $3 trillion in incremental funding by 2028, with over half coming from external sources—both debt and equity. On the debt side, Vontobel Asset Management notes Wall Street projects $250 billion in U.S. dollar bond issuance by these firms this year, rising to $400 billion globally, with more than 40% of that debt having maturities beyond 15 years, reflecting the long economic life of data centers.
This has already squeezed other sectors, particularly construction outside data centers. Residential housing is next, as mortgage rates have climbed above 7%, stoking concerns about a broader slowdown. Yet recent data show manufacturing and services strengthening, and unemployment remains low. A Financial Times survey of prominent bond investors found a prevailing view that soaring yields are "not even close" to cooling the economy.
My own perspective is that yields are returning to normal levels after an abnormally low-growth period following the 2008 financial crisis. The more apt comparison may be the late-1990s internet boom, when telecom investment drove strong growth. The Fed now faces a critical question: is the economy's potential growth rate higher than its current 1.8% estimate, or is the capital crunch transitory? If the former, the long-term real yield assumption must be revised upward.
Finally, the Fed must weigh financial risks, as the AI buildout increasingly relies on external funding. The biggest danger, according to Van Nieuwerburgh, is that return expectations are too optimistic. As long-term Treasury yields hit 20-year highs, the stakes for policymakers are rising. Meanwhile, public sentiment on the economy remains grim, adding political pressure. The outcome will shape both financial markets and the broader economy in the years ahead.
