Scott Bessent just defined market panic—and accidentally diagnosed the biggest problem with AI

“Markets go up and down,” Bessent said. “What’s important is that they are continuous and functioning. In my 35-year career, when people panic is when you’re not able to have price discovery—when markets close, when there is the threat of gating, things like that.”​

It’s a tidy, veteran-investor definition of systemic risk. Volatility, he implied, is fine. Volatility is information. The true crisis arrives when the mechanism that produces prices breaks down entirely—when buyers and sellers can no longer reliably find each other and agree on what something is worth.

Bessent was talking about bond markets and the Strait of Hormuz. But he might as well have been talking about AI stocks (or lack thereof).

The real problem isn’t the selloff

The AI trade has surged and then unraveled in ways that look superficially like a normal correction but feel structurally different. Nvidia posted revenue up 73% year-over-year last quarter and watched its stock fall. The Magnificent 7 is down roughly 7% year to date. DeepSeek rattled the sector in January 2025, and the tremors haven’t fully stopped. On the surface, this reads as a rotation or a valuation reset. Underneath, something closer to Bessent’s definition is at work.​

The problem isn’t that AI stocks are dropping. The problem is that nobody credibly knows what they should be worth—which means price discovery, in any meaningful sense, has been severely compromised for years. And that problem is actually worse than the public market selloff suggests, because the most consequential players in AI have never been subject to market pricing at all.

OpenAI is worth $840 billion—or so its latest funding round implies. Anthropic is valued at $380 billion. xAI at $250 billion. These numbers are not prices. They are negotiated fictions, set in private deals between a small number of investors with massive incentives to mark the sector upward. There is no continuous market, no daily clearing mechanism, no army of short sellers stress-testing the assumptions. There is only the last round, which is whatever the most recent believer agreed to pay. By Bessent’s own definition, this is the condition he fears most: not volatility, but the absence of price discovery entirely.

The tremors are beginning to move downstream. Private credit markets—which rushed in over the past two years to finance AI infrastructure, data center buildouts, and hyperscaler supply chains that traditional bank lenders wouldn’t touch—are sending tremors through markets. Jamie Dimon memorably warned of “cockroaches” in October 2025 when a firm in the space, First Brands, filed for bankruptcy. In February earlier this year, another firm, Blue Owl, rattled markets further by moving to restrict withdrawals. Fortune‘s Shawn Tully warned earlier this month about a potential $256 billion meltdown in the sector.

When the public market begins questioning whether Nvidia’s margins are durable, or whether the $650 billion in projected AI capex actually generates returns, the entire chain of private financing built on those assumptions starts to look shakier. Private credit doesn’t have a ticker. It doesn’t reprice in real time. It reprices in defaults, restructurings, and fund gates—exactly the kind of market event Bessent spent 35 years dreading.

When capital floods a sector on the basis of narrative momentum rather than demonstrated cash flows, prices stop being signals. They become votes. And votes, unlike prices, don’t have to be right. The bill for that distinction, in AI, may be arriving on both sides of the public-private divide at once.

That’s the condition Bessent fears in bond markets: not volatility, but the absence of reliable pricing. AI equities have been living in exactly that condition since at least 2022.

When the crowd is right 85% of the time

Bessent has a framework for this, too—one he shared earlier in the same interview. “The crowd is right 85% or 90% of the time,” he told Frost, describing the macro-investing mindset that made him one of the most successful hedge fund managers of his generation. “It’s really that when things turn, or when you could imagine a different outcome than the consensus, that’s when you can really make a lot of money.”​

He cited his bet against the British pound in the Exchange Rate Mechanism crisis (when he and George Soros helped “break” the Bank of England) and his decade-long short of the Japanese yen—both situations where elite consensus had hardened around a mispricing so obvious in retrospect it seems almost embarrassing. In each case, the problem wasn’t that markets were volatile. The problem was that markets had stopped pricing correctly, then snapped back violently when reality reasserted itself.

That’s precisely the tension AI investors are sitting with now. The question is not whether AI is transformative—it almost certainly is. The question Bessent spent his career asking is the one Wall Street forgot to ask for three years: at what price? And more importantly—is there even a mechanism right now to answer that question honestly?

The Lifeguard’s Lesson

At one point in the interview, Bessent reflected on his teenage years as a lifeguard, offering what he called a lesson that carried into both investing and politics. “Drowning people will try to pull you down,” he said. “many drowning people can just be saved by stand[ing] up,” he added, “so, a lot of times people are panicked, in the water.”​

It’s a striking image for the current AI moment. The next time the market thinks it’s drowning, it could just be panicking in shallow water, thrashing against a depth it can’t measure, precisely because the floor—real, grounded, fundamental value—has never been clearly established. Price discovery doesn’t just tell you what something is worth today. It tells you whether you’re standing or swimming.

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.