Listen to the audio version of this article (generated by AI).
Hello, Reader.
“The biggest and fastest rise and fall in business history.”
No, that’s not a reference to Nvidia Corp.’s (NVDA) terrible week.
It’s The Economist’s description of the 2001 telecom sector crash.
In short, telecom companies borrowed heavily in the late 1990s to build fiber optic networks for the internet boom, but demand didn’t grow quickly enough to repay their debts. Investors poured billions into the sector, driven by unrealistic expectations for growth and profitability.
Eventually, lenders pulled back, and many companies went under, leaving much of the fiber unused for years.
The damage was staggering: In the five years after the 1996 Telecommunications Act, which deregulated the industry and invited a rush of new competitors, telecom companies spent over $500 billion – mostly borrowed – building networks.
By 2002, they owed about $1 trillion. WorldCom and Global Crossing, two of the major U.S. telecommunications giants, went bankrupt… and bond investors recovered just over 20 cents per dollar.
Today, a different kind of network is going up. Instead of fiber, it’s data centers, chips, and the power that runs them. And once again, much of it is being built with borrowed money –before the profits are there to justify it.
And now the lenders are getting nervous.
This week, Reuters reported that some banks, which had recently lined up to fund AI deals, are now reluctant to finance more – even with guarantees from marquee names like Alphabet Inc. (GOOGL) and Anthropic. The shift followed last week’s selloff that wiped $2.5 trillion from the Nasdaq-100 in one day, sending it into correction.
In today’s Smart Money, I’ll explain what’s happening with banks funding the AI boom, which could look like a repeat of the telecoms crash, and what that means for your portfolio.
The Money Behind the Machines
From the start, the AI boom has relied heavily on borrowed money.
Last year alone, Amazon.com Inc. (AMZN), Microsoft Corp. (MSFT), Meta Platforms Inc. (META), and Alphabet collectively poured nearly $300 billion into capital expenditure. That figure will more than double this year to an eye-watering $635 billion.
And as the hyperscalers deplete their cash reserves to build data centers, they are tapping the credit markets for additional financing. Annual issuance of debt tied to AI and data centers surged from $166 billion in 2023 to $625 billion last year.
That appetite for debt is now colliding with caution…
Banks and credit investors are beginning to scrutinize AI deals more closely and becoming more cautious about lending.
Concentration is one reason; they’re already heavily exposed to the sector. At JPMorgan Chase & Co. (JPM), loans backed by income-producing properties, a category that includes data centers, totaled $29 billion at the end of the third quarter. Data centers and similar properties now make up 12% of JPMorgan’s commercial real estate loans, up from 8% three years ago. It’s the bank’s second-largest property exposure.
Chris Poggi, a finance partner at Ropes & Gray, put it this way: Debt investors’ books “have become full,” and they are “assessing if they have extended themselves as much as they’re going to extend themselves this year.”
Yet the need for cash isn’t going away. Goldman Sachs Research projects that global hyperscaler debt issuance will reach $420 billion in 2027, more than 60% above its estimate for 2026.
That is the squeeze: Borrowing needs are climbing just as lenders get more cautious.
This is how money could become the new AI bottleneck. For the past two years, the limits on the buildout were chips and power. If lenders keep pulling back while borrowing needs keep climbing, the constraint becomes capital.
And this could hurt us investors.
Here’s how a pullback in lending could set off a chain reaction:
- Banks become reluctant to lend. Some already are, even for deals backed by blue-chip guarantees.
- AI infrastructure companies face higher borrowing costs. According to Bank of America credit analyst Yuri Seliger, big cloud companies are paying more to borrow money. Their borrowing costs have risen about 0.5 percentage points this year, compared with just 0.2 points for other financially strong companies not tied to AI.
- Data center projects get delayed or scaled back. Some planned bond offerings tied to data centers and chips are already being shelved. Nebius Group N.V. (NBIS), for example, scrapped plans for a bond sale to raise money.
- Demand for chips, power, and construction could weaken. Every delayed project is an order that never gets placed.
- Infrastructure investors face pressure. Companies whose growth depends on that steady stream of orders may see growth slow and their stock prices fall.
That’s why it’s crucial for investors to be more selective in this phase of the AI trade.
Why Selectivity Matters Now
The market is shifting from rewarding anything tied to AI to sorting the strong from those with stretched balance sheets, and the stretched are already feeling it.
The pullback has been sharpest among riskier borrowers, like neoclouds. CoreWeave Inc. (CRWV) is down 33% this year, and its bonds have dropped sharply. So, investors are becoming less confident in the company’s ability to repay its debt, or they’re demanding higher returns to take on that risk.
Now, none of this means the AI boom is over.
The telecom companies of the late 1990s weren’t wrong about the internet. They were wrong about how long lenders would keep paying for it.
The investors best positioned now are the ones asking whether a company has the cash flow and balance sheet to keep building if money gets harder to come by.
We’ve been preparing for this shift at Fry’s Investment Report, focusing on companies with the financial strength to keep growing even when capital gets harder to find.
In my upcoming November issue, due out on Monday, I’ll examine what else lenders are signaling about the AI boom’s staying power – and why those signals deserve attention, no matter how strong AI stocks appear.
I will also detail my newest recommendation, an energy pick built for the post-midterm landscape and the AI boom’s power demands.
To learn how to join Fry’s Investment Report and access the latest monthly issue, click here.
Regards,
Eric Fry
