Semiconductor stocks retreated in premarket trading on Thursday as a renewed bond market selloff pushed the 10-year Treasury yield above 5.1%, raising questions about the sustainability of the artificial-intelligence trade. Nvidia fell 1.3%, AMD dropped 2.3%, and Intel declined 3.4% ahead of the opening bell.
The move came after stronger-than-expected business activity data and rising cost pressures revived expectations that the Federal Reserve may need to tighten monetary policy further. The S&P Global flash PMI for September showed the composite output index jumping to 58.4 from 56.0 in August, the highest reading since July 2021, while input costs accelerated at the fastest pace in nearly four years.
Higher Treasury yields pose a dual challenge for growth-oriented tech companies. They increase the discount rate applied to future earnings, reducing the present value of long-dated cash flows, and they offer investors a risk-free alternative with attractive returns. Emily Bowersock Hill, CEO of Bowersock Capital Partners, told Barron's that "Treasury yields at this level pose significant competition to equities."
The timing is particularly sensitive because semiconductor valuations were repriced aggressively just days earlier. AMD surged 9.9% on Monday, crossing $1 trillion in market value for the first time, while Intel jumped 12.1% and Nvidia gained 2.3%. That enthusiasm was driven by hopes that agentic AI and rising inference workloads would broaden compute demand beyond GPUs to CPUs and other infrastructure.
Now, the yield shock lands on stocks that had just become more expensive. Goldman Sachs' Anshul Sehgal, however, offers a counterargument. In a September 18 interview published by Goldman, he acknowledged that investors are increasingly drawn to long Treasuries above 5% but argued that the better risk-reward remains in compute. "Personally, I think the asymmetric expression is being long compute," Sehgal said, noting that while a Treasury offers a fixed return, successful AI infrastructure could deliver far larger upside if computing demand expands as expected.
So far, there is little evidence that the semiconductor pullback reflects a downturn in AI spending. Hendi Susanto, a portfolio manager at Gabelli Funds, told MarketWatch that customers continue signing multiyear agreements for chips, networking equipment, and other infrastructure. "That's not what the front end of a downturn looks like," he said. Susanto also highlighted inference, enterprise AI, and physical AI as future demand drivers, while noting that semiconductor capacity takes years rather than quarters to add.
That distinction is crucial. Nvidia, AMD, and Intel can fall because investors demand lower multiples even while customers continue ordering more compute. If Treasury yields remain above 5%, Wall Street may become less willing to reward every company attached to the AI theme equally. Investors will increasingly demand proof that capital spending converts into revenue, margins, and cash flow quickly enough to beat a risk-free alternative.
This article is for informational purposes only and does not constitute financial advice.
