Cramer: AI stocks untradeable as data center uncertainty wipes out valuations

The iShares Semiconductor ETF (SOXX) fell 3% on Wednesday and another 10% on Thursday, and CNBC's Jim Cramer says the reason is simple: nobody knows what AI stocks are worth anymore. The host of "Mad Money" told viewers Thursday that uncertainty over the future of the data center build-out has made artificial intelligence stocks increasingly difficult to value, producing steep selling across two brutal sessions.

The sell-off did not discriminate. It swept across the entire AI trade, hitting everything from chipmakers to the companies that supply power and electrical equipment for data centers. GE Vernova, the power equipment maker whose fortunes are tied to electricity demand from massive computing facilities, dropped 7% on Thursday. Sandisk, the memory and storage company, fell 22% the same day. The numbers describe a broad repricing, not a targeted correction in a single name.

What is driving the selling?

Investors are worried about a specific mechanism, Cramer explained. Regulatory hurdles and local opposition to new data centers could slow the build-out of the physical infrastructure that AI systems require. That, in turn, would undermine the growth forecasts on which chipmakers' current valuations rest.

The logic chain matters for anyone holding these stocks. Chip companies command premium multiples because analysts expect demand to remain strong for years. If regulators refuse to permit new data centers at the pace the industry currently enjoys, the demand assumptions break. When the assumptions break, the valuations built on them lose their foundation.

Cramer put the problem in blunt terms: "We don't know what to pay for any of the stocks connected to the data center and that's why this moment is so fraught."

That sentence is the heart of the matter. It is not a claim that the companies are weak. It is a claim about the limits of analysis itself — when the future permitting environment for data centers is unknowable, standard valuation models produce answers nobody trusts.

Why Nvidia looks cheap but may not be

Cramer pointed to Nvidia as the prime example of the valuation trap. The AI chipmaker trades at roughly 12 times next year's earnings, with an expected growth rate of 70%. On its face, that combination looks remarkably inexpensive — a company growing 70% a year, priced at a multiple more often associated with slow-moving industrial firms.

"That's an incredible bargain," Cramer said.

Then came the caveat, and it is the caveat that explains the market's fear. Those earnings projections assume demand will remain strong for years to come. The multiple is only cheap if the earnings arrive.

"But it is not a bargain if, the following year, it can't grow any further, because the U.S. won't permit new data centers as fast as they do now," Cramer said.

The structure of the argument is worth pausing on. A 12 times forward multiple with 70% expected growth would ordinarily signal either a market error or a hidden risk. Cramer identified the hidden risk as regulatory: the permitting pipeline for data centers, not the technology, not the competition, and not the demand from customers who want the chips today.

How exposed is Micron?

Memory maker Micron presents a different version of the same problem. Micron's long-term take-or-pay contracts — agreements in which customers commit to paying for chips whether or not they take delivery — provide some revenue visibility even if customers do not accept their orders. In theory, that contractual structure insulates the company from a demand air pocket.

Cramer, however, said he is unsure how sturdy those contracts would prove if any signees fell on especially hard times and did not have the money to pay. A contract is only as strong as the counterparty behind it. If the data center slowdown squeezed Micron's customers financially, the take-or-pay provisions that currently comfort investors could turn out to be worth less than the paper they are written on.

The distinction matters for how investors read Micron's revenue guidance. Contracted revenue and collectible revenue are not the same thing, and Cramer's skepticism centers precisely on that gap.

What about AMD?

Advanced Micro Devices offers the third variation on the theme, and it is the one Cramer framed as a genuine dilemma. He praised AMD's latest earnings report in unusually strong language, calling it "one of the best quarters I have ever seen." He also highlighted CEO Lisa Su's projection that the broader data center market could eventually reach $1 trillion.

"If that's the case then you can buy her [CEO Lisa Su's] stock right now," Cramer said.

But he immediately turned the projection against the price. "Do you want to? The stock was at $188 in March. By that token, $404 seems awfully high to me. Maybe that's the real problem."

The numbers carry the argument. AMD has more than doubled from $188 in March to $404, and even a $1 trillion total addressable market projection from the CEO does not automatically justify every dollar of that increase. Cramer's point is not that AMD's business is faltering — he called the quarter one of the best he has ever seen — but that the stock price already embeds enormous success, and possibly more than enormous success.

Delay or derailment?

Cramer distilled the entire situation into a single question: will regulatory hurdles significantly slow data center construction, or merely delay it? The difference between those two outcomes is the difference between impaired growth and deferred growth, and the market currently cannot tell them apart.

Until the outlook becomes clearer, even exceptional earnings outlooks may not be enough to justify AI stocks' valuations, Cramer concluded. That is the sobering conclusion for investors scanning the wreckage of a two-day sell-off: the problem is not the reported results, which in AMD's case Cramer described in superlative terms. The problem is the discount rate applied to an uncertain future, and no earnings report can resolve a question about whether the United States will permit new data centers at the current pace.

Why the disclosure matters

Cramer's Charitable Trust, the portfolio run by CNBC's Investing Club, owns shares of Nvidia (NVDA) and Micron (MU), the show disclosed. Readers weighing his analysis should note that the commentator holds positions in two of the companies whose valuations he was assessing — a fact relevant to interpreting the tone of his remarks, which criticized prices while praising the underlying businesses.

What comes next

The immediate question for markets is whether the two-session rout marks a repricing or a turning point. Cramer's framework suggests investors should watch one variable above all: the pace of data center permitting and construction in the United States. If approvals continue at the current rate, the 70% growth assumptions behind Nvidia's 12 times multiple regain credibility and the sell-off looks like an overreaction. If approvals slow, the take-or-pay contracts at Micron and the $1 trillion projection at AMD both become harder to underwrite. Until that evidence arrives, Cramer's warning stands: nobody knows what to pay for any stock connected to the data center, and markets that cannot price assets tend to keep selling them.