The surge in U.S. bond yields to two-decade highs has raised questions about what is causing it, how long it will last and how it will impact the U.S. economy. Although inflation has been well above the Fed’s 2 percent target this year, the rise in bond yields since late August has not been associated with an increase in inflation expectations. Instead, it is mainly attributable to a rise in real yields.

Fed Chair Kevin Warsh cited three forces at play in his September Federal Open Market Committee press conference, including underlying economic strength, surging capital spending and geopolitical factors. The first two of these are related to the unprecedented buildout in artificial intelligence infrastructure.

I previously highlighted how the boom in AI infrastructure was supplementing consumer spending as a driver of U.S. economic growth. The current estimates of capital spending for five hyper-scalers — Amazon, Alphabet, Meta, Microsoft and Oracle — call for it to surpass $800 billion this year and $1.1 trillion next year.

Professor Stijn Van Nieuwerburgh of Columbia Business School puts the prospective AI capital expenditure spend into historical perspective. He estimates that the pipeline of investments being contemplated would entail more than $10 trillion from 2025 to 2032, representing 3.6 percent of GDP per year. If so, it would far exceed the respective infrastructure buildouts for canals, railroads, electrification, highways or telecommunications.

One caveat is that calculations of AI capital expenditure to GDP overstate the overall economic impact, because a substantial portion will leak out as increased imports. The effect on economic growth thus far is mainly driven by increased business capital spending. However, as more businesses adopt AI and productivity increases over time, it will become more ingrained.

A study by the Federal Reserve Board of Governors finds that adoption trends thus far provide a mixed picture of the ubiquity of AI. The evidence overall is consistent with a buildout phase rather than the onset of broad-based displacement: “While some highly exposed sectors show relatively strong productivity, labor market impacts remain concentrated and have not yet broadened in the aggregate.”So, how are these developments impacting U.S. bond yields?

The answer is that the volume of capital spending by the hyper-scalers is so large now that the financing requires more than internal cash flows can generate. Morgan Stanley estimates that more than half of the $3 trillion required to meet their incremental requirements over 2025-2028 will come from outside capital. The funding sources include both debt financing and third-party equity.

On the debt side, Swiss-based Vontobel Asset Management reports that Wall Street estimates point to $250 billion of U.S. dollar bond issuance by hyper-scalers in the U.S as of year-end, and $400 billion globally. Beyond this, the report estimates the five companies have placed more than 40 percent of their issuance beyond 15 years. The rationale for issuing long-term debt is that data center assets have long economic lives.

This shift in private sector financing is occurring while the federal budget deficit is an estimated 6 percent of GDP and federal debt outstanding held by the public is approaching a record high. As a result, there is now increased competition for capital.

Various sectors of the economy are feeling the squeeze, especially the construction sector, excluding data centers. Looking ahead, residential housing is also likely to be impacted as mortgage rates have surged above 7 percent. This has fueled a debate about whether rising yields will curb overall economic activity or co-exist with strong economic growth.

What is playing out thus far is consistent with the view that the AI investment boom is powering the economy. Recent economic data, for example, show a strengthening of both the manufacturing and service sectors of the economy and continued low unemployment. A Financial Times article surveyed prominent bond investors, and the prevailing view was that soaring bond yields were “not even close” to cooling the U.S. economy.

My own take is that bond yields are back to more normal levels after U.S. economic growth was abnormally low in the aftermath of the 2008 financial crisis. The more relevant comparison is with the internet boom in the second half of the 1990s when the telecom sector was a key driver of strong economic growth.

Amid this, the Federal Reserve must assess whether the economy’s long-term potential growth rate is above its current estimate of 1.8 percent, or whether the competition for capital is a transitory phenomenon. If the former is true, its calculation of the long-term real bond yield would need to be adjusted higher.

Finally, the Fed also needs to assess the financial risks, as the AI buildout is increasingly reliant on external sources that Professor Van Nieuwerburgh cites. The biggest risk, however, could be that return expectations for the buildout are too high.

Nicholas Sargen, Ph.D., is an economic consultant and is affiliated with the Darden Business School. He has authored three books including “Investing in the Trump Era.”