Director Lee Eun-taek of KB Securities pointed to U.S. long-term interest rates as the key variable that will determine the next upcycle in the stock market. If rates rise to levels the market is not used to, capital allocators such as banks and pension funds could shift funds from risky assets to Government Bonds, adding pressure to stocks.
Lee said at a press briefing at the Korea Exchange (KRX) on the 18th, "I think it could be dangerous if the U.S. 10-year Government Bonds yield trendwise breaks into the low 5% range," adding, "If it goes above 5%, it is the record high since the 2007 financial crisis, and if it exceeds 5.3%, it is the highest level in 25 years." The 10-year yield previously climbed to 5.4% in 2002, 5.3% in 2007, and 5.0% in 2023.
The crux is that rising rates can even change the "money taps" of the asset market. If U.S. Government Bonds yields rise to high levels, institutional investors such as pension funds have less incentive to take greater risks by supplying capital to AI corporations or data centers. That is because they can secure high returns with relatively safe Government Bonds alone.
However, he also stressed that the stock market bubble does not collapse just because rates rise. KB Securities analyzed 120 years of market history and found that although the backgrounds of major bubble bursts such as 1929, the late 1960s, and 2000 differed, trendwise rate increases commonly preceded them.
In particular, as criteria for judging dangerous rate increases, he presented "No way back" and "Breaking to new highs." The point is that a situation where tightening becomes hard to reverse due to inflation is more dangerous than a scenario where rates jump temporarily because of war or a spike in oil prices and then fall again. He also noted that when rates break to new peaks for the first time in 10 to 20 years, rather than levels the market has already experienced, investors' behavior itself can change.
In the artificial intelligence (AI) investment cycle, he also cited the movements of capital suppliers rather than big tech itself as the key variable. Hyperscalers such as Meta and Amazon can capture outsized revenue if their businesses succeed even as they continue massive investments, so the likelihood they stop investing on their own is low.
By contrast, capital suppliers such as banks, pension funds, and sovereign wealth funds have a different structure. Even if a project succeeds, it is hard for them to earn beyond contracted revenue, but if it fails, they can suffer principal losses, making them far more sensitive to corporations' cash flows and repayment capacity.
Lee said, "If you focus on whether big tech will invest more or not, it can be hard to get the answer right," adding, "A much better question is what conditions would make capital suppliers stop." He also assessed the recent rise in Oracle's credit default swap (CDS) premium as a case that revealed capital suppliers' anxiety.
As another backdrop to the recent stock market correction, he cited weakening demand for frontier AI models. As corporations become conscious of expense burdens, they have shifted from "token maxing," which uses as many tokens as possible on the highest-performing AI models, to "token optimization," which uses high-performance models only for necessary tasks.
However, Lee saw that concerns about demand for such frontier models may have been largely reflected during the recent share-price adjustment. While frontier model companies are cutting prices, Chinese AI companies that had engaged in low-price competition are starting to raise prices to secure profitability.