Why AI Stocks Got Hammered Wednesday

I’ll be back next week to finish up my series on the AI stack. But Wednesday’s selloff deserves its own conversation, so let’s talk about what sank AI stocks.

It started Tuesday, when investor Chamath Palihapitiya posted that the AI building boom might be inflating a private credit bubble, and Elon Musk answered with one word: “Yikes.”

Then Wednesday’s tape turned bloody, with CoreWeave (CRWV), the biggest neocloud renting out AI computing power, down 15.5%. Nebius (NBIS) lost 10%. Data center developers TeraWulf (WULF), Cipher Digital (CIFR), and Iren (IREN) each fell more than 12%. Oracle (ORCL) dropped almost 9%, and Dell Technologies (DELL) slid more than 11%.

I’d love to point the finger at a single villain, but I think three things went wrong at once. Money got more expensive, lenders got nervous, and Wall Street started asking whether the customers are really there.

All three circle back to the same question.

Who’s paying for all of this, and what will it cost them?

Let’s start with the price of money, because that hit first. An hour before the opening bell, the September inflation report came in hot, with core prices up 0.3% for the month against forecasts of 0.2%.

A tenth of a percent doesn’t sound like much, but before the report, futures markets priced in a 22% chance of a Fed rate hike at the October 28 meeting. By the afternoon, they were close to a coin flip.

We’ve talked about how the AI buildout runs on borrowed money. So, think about building a house with a construction loan. You pay interest long before a tenant moves in, and if rates jump halfway through, the house is worth the same, but your return shrinks. That’s the world data centers and neoclouds are now navigating.

That brings us to the second problem — nervous lenders.

Back in June, CoreWeave sold bonds due in 2032 that pay 9.625% a year. Buy $1,000 worth, and you collect $96.25 a year. On Wednesday, those bonds traded below 80 cents on the dollar. A new buyer paying around $800 still collects the same $96.25, about 12% on their money. But they also get the full $1,000 back when the bonds come due in 2032, and once you add that extra $200, the return works out to about 15% a year.

CoreWeave isn’t paying 15% on what it already owes, because those rates are locked in. The worry is the next loan, because that’s where lenders are pricing the risk now. And CoreWeave’s interest bill already ran $640 million last quarter, close to a quarter of its revenue.

CEO Mike Intrator told employees the market was acting irrationally and that the company’s $104 billion backlog more than supports its debt. I get it. That’s an impressive number of signed contracts. But from the market’s bearish lens, customers don’t pay until the computing power is up and running, and CoreWeave has to borrow the money to build it first. That need to borrow billions has spooked investors, lenders, and suppliers.

Then Gartner piled on, predicting that cost pressure will force more than 80% of neoclouds to give up their independence by 2030. When most investors hear a company might get bought, they picture a nice fat premium. Gartner’s talking about companies that run short on cash and viable alternatives, and get folded into somebody bigger on the buyer’s terms. I don’t think Gartner’s prediction reflects an industry-wide belief, but with CoreWeave’s bonds already sliding, it was the last thing anyone wanted to read.

And then there’s Dell, which suffered a headshot unrelated to interest rates or CoreWeave. Barclays analyst Tim Long downgraded Dell, Hewlett Packard Enterprise (HPE), and NetApp (NTAP), and he didn’t sugarcoat why. “Valuations have run, commodity costs have helped revenues, and we don’t see a meaningful AI inflection yet.”

If sentiment weren’t in the gutter, this downgrade likely doesn’t get much airplay. But on a day like Wednesday, it was like pouring gas on an already out-of-control engine fire.

Dell nearly quadrupled over the past year, so yeah, expectations were off the charts. But Barclays figures the average company spends only about 5% of its IT budget on AI, and only about a quarter of that goes to hardware. And when the memory inside a server costs more, Dell charges more for the same server, so sales go up even if it isn’t shipping more boxes.

Once you see how these companies are tied together, the rest of the day makes sense.

Developers are basically landlords, and the neoclouds are often their tenants. If a tenant suddenly has to pay 15% to borrow, the landlord starts sweating, and so does everyone selling gear to the building.

Meanwhile, Microsoft (MSFT), Alphabet (GOOGL), and Amazon (AMZN) each lost less than 2%, arguably, because they can pay for their data centers out of pocket. As simplistic as this sounds, Wednesday’s AI selloff felt more like the market sorting companies by who can afford the buildout without leaning on a lender, and who can’t.

The news on Wednesday wasn’t all bad. ASML (ASML), which makes the machines that print computer chips, posted record quarterly sales and said some customers already want to place orders for 2029 and 2030, though its shares still fell almost 4%.

BlackRock CEO Larry Fink said his firm helped finance Meta’s planned $14 billion data center in Texas, so the money’s still out there for the right deals.

So Where Does That Leave Us?

Before Wednesday’s inflation report, the odds favored the Fed sitting tight on October 28. Now it’s a coin flip. That same week, the big cloud companies report earnings, and if they’re still spending like drunken sailors on AI, I’m not saying the punch bowl comes right back out, but a lot of the worry about who’s paying starts to fade.

I want to be clear that I’m an optimist about AI, and demand isn’t the problem. The current problem is the cost of money, and whether these companies can line up financing and still deliver a strong return to shareholders. Nobody knows the answer yet, and that’s really what this whole fight is about.

Markets read the news through whatever mood they’re in. When investors are bullish, everything looks great through rose-colored glasses. When they’re bearish, what a company says almost doesn’t matter, and that’s where we are right now. Cipher Digital raised $850 million on Wednesday without paying a dime of interest, and its stock still fell 12%. Instead of talking about how fast AI is growing, everybody’s talking about what it costs.

Many of these stocks are below their 50-day and 200-day moving averages, which means they’re in downtrends, and most investors aren’t interested in catching a falling knife. At some point, longer-term investors will see enough value to step in, or something will change their minds. Until then, I expect traders to wait for these stocks to hold their gains on good news or stop falling on bad news, rather than guessing the bottom.

More From Bob Byrne’s AI Series:

At the time of publication, Byrne had no positions in any securities mentioned.