Investing

AI Slowdown Fears Hit Chips as the 10-Year Treasury Yield Finally Broke 5%

The S&P 500 fell just 0.48% while semiconductors collapsed 5.86%. What matters is not how much the index dropped, but what the market chose to sell first: the AI-pace debate hit chip earnings expectations, and the 10-year Treasury's push past 5% raised the discount rate applied to those expectations.

사진 Thomas J. O'Halloran · Public domain · 위키미디어 커먼즈

The index was calm, but a storm hit inside semiconductors

On September 14, the US stock market looked, on the surface, like an ordinary pullback. The S&P 500 fell 0.48%, the Nasdaq fell 0.56%, and the Dow fell 0.29%. But on that same day, the Philadelphia Semiconductor Index dropped 5.86%. Nvidia fell 3.36%, AMD fell 4.40%, Broadcom fell 4.77%, and Micron fell 5.25%.

  • -0.48%: The broad market saw a sub-1% pullback
  • -5.86%: A sharp repricing in semiconductors, where AI infrastructure expectations are concentrated
  • Above 5%: Intraday, the first time since 2023

Meanwhile, ServiceNow, Adobe, and Workday, all of which had been under pressure lately on fears of AI replacing their products, rebounded between 4% and 7.4%. That pairing is fairly telling. The market did not decide "AI is bad." It recalculated who stands to lose first if AI slows down a little, and who actually buys themselves more time.

If training speeds slow, the numbers most likely to wobble first belong to GPUs, high-bandwidth memory, high-speed networking, servers, and related equipment. For established software companies, on the other hand, a slower pace of frontier-AI development can mean their own products get replaced more slowly. One day's price action does not prove the industry's structure has changed, but the direction money moved was unmistakable.

Why chips got hit harder than software

The core of the AI-slowdown argument is not a call to stop using AI. Leading AI executives, including Anthropic's Dario Amodei, have argued that the gap between how fast frontier models are improving and how thoroughly they are safety-tested needs to narrow. Once that argument surfaced, the market began, for what looks like the first time, connecting it directly to the timeline for chip demand.

Semiconductor stock prices are not explained by current earnings alone. They price in years of bigger models, more training, more inference, and rising data-center capital spending to match. So a six-month delay in the development timeline does not erase long-term demand, but it can push back when that revenue actually shows up and how long it takes to pay back the investment. At today's high valuations, that timing gap matters.

This sell-off looks less like "AI is over" and more like the market asking, "will AI infrastructure revenue arrive as fast as we assumed?"

Notably, even after this drop, the Philadelphia Semiconductor Index (SOX) was still up roughly 57% for the year. The more expectations an asset has already priced in, the more sensitive it is to fresh uncertainty. Reuters Breakingviews also noted that this move sits within the range of past volatility. In other words, a single day's 5-6% drop is not yet enough evidence to declare a structural collapse in demand.

A 5% 10-year Treasury yield is the second shock

On the same day, the 10-year Treasury yield crossed 5% intraday, the first time since 2023. It did not close the day above 5%, though, and that distinction matters. The 5% level is not yet a new floor; it is closer to a boundary the market is testing again for the first time.

Still, it matters a great deal to growth-stock investors. A stock's value is calculated by discounting future cash flows back to the present. As the 10-year yield approaches 5%, companies expected to generate large profits far in the future feel the impact of a higher discount rate the most. Tesla, AI software names, and high-P/E chip stocks are the clearest examples.

This time, chips were hit from the worse angle. The AI-slowdown debate struck the numerator, future earnings expectations, while the 5% 10-year yield pushed up the denominator, the discount rate. Both pressures hit valuations from opposite directions on the same day.

Demand-side pressure The possibility that model development and data-center buildouts get delayed. No order cancellations have been confirmed yet, but uncertainty has been added to the earnings timeline.

Discount-rate pressure Near a 5% 10-year yield, investors may accept a lower P/E for the same earnings outlook. The higher the growth rate a company promises, the more it now has to prove it.

The market did not give up on AI; it recalculated when AI revenue will actually show up.

Insight Times Editorial Desk