[Insight & Opinion] Can AI Beat Interest Rates This Time, Too?

AI Boom Faces the "High-Interest-Rate Test"

Surviving on Borrowed Money Could Lead to a Fall

The "Price of Money" Will Determine Winners and Losers

On September 30, the U.S. 10-year Treasury yield topped 5.3% intraday, reaching its highest level since 2002. Wall Street analysts shrugged. They believed AI would be different: no matter how high interest rates climbed, the AI boom would continue. This is a belief that has resurfaced with every bubble.

Will that belief prove right this time? Democrats scored a major victory in the U.S. midterm elections last Tuesday, November 3, and the Nasdaq plunged 5.52% on Thursday amid concerns about regulations on AI data centers. It is still too early to answer whether the belief is justified. But we can compare the evidence behind both sides of the debate side by side, in numbers. Of course, we should keep in mind that the midterm election results so far do not fully reflect the impact on AI.

First, the case for AI winning. One, this time there is actual revenue. Third-quarter cloud revenue rose 103% at Google, 48% at Microsoft Azure, and 39% at Amazon Web Services. Microsoft's commercial contract backlog, in particular, grew 73% to $677 billion. That is a very different starting point from the dot-com era, when valuations rose on expectations alone. Two, productivity. U.S. nonfarm productivity rose 1.4% at an annualized rate in the second quarter of this year and was up 2.2% year over year, which economists interpreted as an effect of companies adopting AI. Federal Reserve Chair Kevin Warsh also said at a meeting held last month, "Growth driven by productivity does not necessarily fuel inflation." If productivity takes off, growth will lower the debt ratio and interest rates, justifying today's investment. Three, there are companies with the financial strength to withstand the pressure. The U.S. economy itself is also solid: second-quarter growth was revised up to an annualized 2.2%, and consumer spending to 3.8%. Microsoft generated $58 billion in operating cash flow in the third quarter, leaving it with $18.6 billion in free cash flow even after making enormous capital investments.

Leading the opposing case—that AI will lose—is Ruchir Sharma, chairman of Rockefeller International. In a column published in the Financial Times, he argued that "the AI boom cannot withstand higher interest rates." He offers three reasons. One, this investment is being financed with debt, not cash. According to FactSet, a global financial data provider, the share of capital expenditure funded by increases in debt at Alphabet, Amazon, Meta, Microsoft, and Oracle jumped from 9% in fiscal 2024 to 32% on a trailing 12-month basis through the middle of this year. In the third quarter, free cash flow was negative $9.1 billion at Alphabet, negative $2.1 billion at Meta, and negative $33.5 billion at Amazon. Sharma expects AI-related debt issuance to exceed $600 billion this year, 10 times the 2024 level. Two, the weakest links are the first to crack. The yield on bonds issued by CoreWeave, a fledgling cloud provider, and maturing in 2032 rose as high as 14.8% annually on October 14. Credit rating agency S&P rated CoreWeave speculative grade (B+). The gap is wide even when compared with the 7.24% yield on Oracle bonds maturing in 2032. Three, interest rates have risen rapidly. While the rate level is similar to that of the 1990s boom, this time rates have climbed sharply over a short period after nearly 25 years of ultra-low rates. The Fed raised its benchmark rate to 3.75%-4.00% in September and held it steady in October, but two policymakers argued for a hike. Markets put the probability of a December rate hike at 78% (as of November 6, local time). Government debt is also at a record high, leaving companies competing with the government for the same pool of money. Even as money pours into AI, it is not spilling over into other sectors. Borrowing is weak and investment is declining outside AI, making it difficult for AI companies alone to thrive when potential customers are struggling, Sharma argues.

How will the contest actually play out? Rather than interest rates defeating AI as a whole, they are more likely to separate the winners from the losers within the AI sector. As Sharma points out, the end of a bubble is not a single moment but a process. In 2000, stock prices fell sharply early in the year, but investors continued pouring money into technology funds through year-end, and the frenzy did not cool until 2001. This time, too, the shock of higher rates may first hit the many companies relying on debt rather than the few with cash reserves, particularly fledgling companies with weak credit. AI will not disappear even if rates rise. What will change is who survives to reap the rewards. The borrowing costs of new companies, free cash flow at big tech companies, and the Fed's December decision are signals to watch in the months ahead.

South Korea is not exempt from this question either. Thanks to the AI semiconductor boom, our total revenue is set to rise by more than 30% next year, and the country has set aside 162 trillion won in additional tax revenue generated by the boom in a Future Response Fund. That shows just how heavily we rely on semiconductors and AI. The Bank of Korea also raised its benchmark rate twice, in July and August, bringing it to 3%. If AI investment falters under the pressure of higher rates, exports and tax revenue could both come under strain. If spending is built on tax revenue from a boom, then when the boom fades, only the spending will remain.

What we need is not faith that AI will win, but preparation for the possibility that it will not. The government should not treat boom-driven tax revenue as guaranteed income, and companies should examine how much of their AI investment they are financing with debt. The price of money has always mattered, and the bill always arrives late. There is no reason to assume AI will be an exception.

Cho Won-kyung, Professor of Economics at Sejong University

This content was produced with the assistance of AI translation services.