The trade that defined 2025—sell bonds, buy anything touched by artificial intelligence—is reversing with startling speed. U.S. Treasuries rallied sharply on Thursday as investors fled the semiconductor stocks that have powered markets for two years, pushing the 10-year yield down more than 15 basis points in a single session. The move suggests something more significant than a routine rotation: the market is beginning to price in the possibility that the AI spending bonanza has overshot reality.
The catalyst was familiar—another round of earnings guidance suggesting that the billions pouring into data centers may take longer to generate returns than Wall Street's models assumed. But the response was different. Rather than buying the dip, institutional investors sold into strength, triggering the kind of correlated decline across chip stocks that typically marks the end of a momentum cycle, not a pause within one.
The numbers tell the story
Global semiconductor stocks shed roughly $400 billion in market capitalization over the past three sessions. Nvidia, the avatar of the AI trade, has fallen more than 12 percent from its September high despite reporting results that would have seemed miraculous to any analyst in 2023. The problem is no longer whether the technology works—it plainly does—but whether the companies building it can earn adequate returns on the capital they are deploying. Hyperscalers have committed more than $200 billion to AI infrastructure this year alone, and the payback period keeps stretching.
Meanwhile, the Treasury market is absorbing the fear. The 10-year note yield, which touched 4.8 percent last week amid stubborn inflation data, has retreated to 4.55 percent. Demand at this week's auctions was robust, with foreign central banks and domestic institutions both stepping up. The bid-to-cover ratios suggest genuine appetite, not just short-covering.
What changed
For most of the past eighteen months, Treasuries and tech stocks moved in the same direction—both sold off when inflation fears rose, both rallied when rate cuts seemed imminent. That correlation is breaking down. Investors are now treating government bonds as a hedge against equity losses, a return to the pre-pandemic playbook that many assumed was dead.
The shift reflects a quiet reassessment of where risk actually lives. AI infrastructure spending has been funded largely by retained earnings and cheap debt issued when rates were lower. As refinancing costs rise and revenue growth disappoints, the leverage embedded in the system becomes more visible. Treasuries, by contrast, offer a guaranteed nominal return in a world where guarantees are scarce.
Our take
This is not the end of the AI story—the technology is too transformative for that. But it may be the end of the phase where investors could buy anything adjacent to large language models and assume the market would bail them out. The flight to Treasuries is a vote for patience over momentum, a recognition that even revolutionary technologies must eventually justify their valuations. For bond investors who spent two years watching their portfolios bleed, the vindication is overdue.
