Jim Cramer urged investors to stop using borrowed money for artificial intelligence stocks as U.S. margin debt climbed above $1.5 trillion.
A lot of the biggest advances in the industry have been in AI infrastructure. Jim Cramer thinks investors who are borrowing to pursue those returns are taking an unacceptable risk.
The CNBC "Mad Money" host urged viewers to get out of data-center investments bought on margin, saying the group is now too dangerous to buy on margin because of the uncertainty about how much AI spending will last.
"If you're borrowing money to buy something related to the data center, then tomorrow morning, 9:30 a.m., sell it no matter what," Cramer said on the July 27 edition of "Mad Money." "You won't regret it."
His warning comes as investors have leveraged up even more against their brokerage accounts.
The Financial Industry Regulatory Authority said debit balances in customer securities margin accounts hit $1.502 trillion in June. That was up 49% from $1.008 trillion a year ago and up 6.1% from May.
The surge doesn't mean all the borrowed money went into AI stocks. But it does indicate that investors have considerably greater leveraged market exposure at a time when data center shares are becoming less predictable.
For shareholders, such a combination produces a dangerous feedback loop. Falling prices might lead to forced sales, which can then drive prices lower and lead to further margin calls.
Cramer's message is therefore narrower than "sell technology." Those investors who possess high-quality equities without leverage are able to weather a slump. Those who are borrowing to enhance their exposure may have little say in when or how their positions are sold.
Buying on margin allows investors to buy more stock than they could if they just used cash.
Under Federal Reserve Regulation T, brokerage companies are normally permitted to loan customers up to 50% of the purchase price of an eligible stock. FINRA rules generally mandate that investors maintain equity in their accounts equal to at least 25% of the value of long securities in the account.
Some brokerage businesses may have more stringent standards, such as maintenance levels of 30% or 40%. They can potentially increase those requirements in moments of market instability.
Related: Alphabet's biggest AI fear may be fading
This means that an investor can get a margin call without making any more trades.
The Securities and Exchange Commission warns that brokers can sell stocks without talking to the customer, and investors may not be able to determine which holdings will be sold. Losses can sometimes exceed the initial investment.
"If you're on margin, get off it," Cramer said. "I no longer feel that you'll get out alive."
That rhetoric seems dramatic, but the math underneath it is simple.
Margin debt reached $1.502 trillion in June.
Customer margin borrowing increased 49% from a year earlier.
Falling stock prices can trigger forced selling.
Brokers may raise maintenance requirements without advance notice.
Investors can lose more than their original cash investment.
An investor who purchases $20,000 of shares with $10,000 of his own funds has used borrowed money to double the size of the holding. A 25% drop takes $5,000 off the stock's value, erasing half of the investor's original ownership position before interest expenses.
Cramer advised investors to seek out companies that gain from data-center expansion but are not fully dependent on the AI boom.
He pointed to building-materials company CRH (CRH) as one example.
Workers just sent AI companies an ultimatum
Palantir CEO has a blunt verdict on OpenAI and Anthropic
Elon Musk pulls no punches with AI rivals as Grok 4.5 debuts
CRH sells materials and infrastructure used in data center projects, but its business also serves highways, bridges, water systems, and other building markets. That diversity might provide some cushion if tech firms reduce their data-center spending.
CRH believes 85% of U.S. data centers are within 25 miles of a CRH facility. At the end of 2025, it was working on more than 100 U.S. data-center projects and was also a leader in roads, aggregates, and water infrastructure.
That doesn't mean CRH is immune to an AI retreat. It does suggest that data-center building is one demand source, not the company's whole investment rationale.
CRH reported first-quarter revenue of $7.37 billion, up from $6.76 billion a year ago. Adjusted earnings before interest, taxes, depreciation, and amortization increased to $586 million from $495 million.
Management reiterated its full-year guidance for net income of $3.9 billion to $4.1 billion and diluted earnings of $5.60 to $6.05 a share. The corporation pointed to demand supports such as infrastructure investment and ongoing reindustrialization activity.
CRH is due to publish second-quarter results July 30 before the market opens, offering investors a new look at whether data-center and public infrastructure demand remains robust.
Cramer is not saying that investment in artificial intelligence is going to go away. The warning is simply a signal that the risk in the deal is shifting.
Individual equities may decline as expectations were too high, valuations got too big, or investors began to doubt how quickly AI projects could pay off, but data-center spending may continue to rise.
Investors who bought the shares outright are able to ride out that volatility.
That might not be an option for investors trading on margin, however. A broker could request additional collateral, increase its requirements, or liquidate shares amid a slump.
That risk is particularly relevant given the 49% yearly rise in margin debt. More leverage can enhance market gains. But it can also turn a normal decline into forced selling.
Cramer's alternative is not to reject technology altogether. It's locating companies with a number of demand drivers and getting borrowed money out of places where prices can fluctuate dramatically.
The difference can make a difference.
The AI boom doesn't have to end for leveraged investors to get crushed. It just needs to be turbulent enough for their brokers to move before the market turns.
Related: Anthropic clarifies stance on open-weight AI models
This story was originally published by TheStreet on Aug 1, 2026, where it first appeared in the Investing section. Add TheStreet as a Preferred Source by clicking here.