Hedge funds dumped tech last week at a decade-record pace.
Goldman Sachs tracks their trades, and their net leverage, the borrowed money they use to juice their bets, fell to a one-year low. When you have to cut borrowed money, you sell what you own the most of, and this year, that’s AI.
Meanwhile, Nvidia is expected to grow its earnings about 70% next year, and it now trades cheaper than the S&P 500.
When forced sellers dump a great business, that’s where the money gets made.
Below, I’ll show you the 2 prices that tell you when to buy the AI selloff and when to get out, and why Meta’s report tonight could hand you the first entry.
What A Real Bubble Needs
People throw the word bubble around a lot, so let’s define it. A bubble needs 4 things: a crash driven by valuations or debt, supply running way ahead of demand, job losses and a recession, and prices that go down and stay down for years.
Think about the dot-com crash. It took the Nasdaq about 15 years to get back to its 2000 high.
A 20% or 30% drop that recovers is a correction. And if you called a bubble the whole way up while the market compounded better than 15% a year for 3 years, a pullback doesn’t make you right.
Now run today’s market through that test.
Everybody isn’t in. Every bubble in living memory, 1987, 2000, the housing peak, came with consumers feeling great. Last month, consumer confidence fell to its lowest level since 2014, and the AI trade is mostly held by the wealthiest investors in the world, who don’t panic because the S&P is 2% off its high.
Supply isn’t ahead of demand. In a bubble, there’s too much of the thing everyone’s building. The rental prices for AI chips have been climbing since mid-September, and memory chip prices made another leg higher.
The debt is small and it’s young. Until last year, the AI buildout was paid for mostly with cash the big companies generated. The neoclouds, the companies that borrow to build data centers and rent out the computing power, carry roughly $60 billion to $70 billion of debt, and most of it is backed by the chips and the contracts.
Compare that to commercial real estate, a $20 trillion-plus market. It went through a pandemic, the fastest rate hikes in 40 years and the Silicon Valley Bank failure, and people called its collapse three times. There was no contagion.
There’s a backstop. When Silicon Valley Bank failed in 2023, the Fed set up an emergency lending program within days. AI is a national-security industry now, and a disorderly unwind of it would get the same treatment.
Nvidia Keeps Getting Cheaper
What the market is doing instead is shrinking the multiple.
A multiple is how much investors pay for each $1 a company earns. A stock at 20 times earnings costs $20 for every $1 of profit.
During the first “AI bubble” scare in 2023, Nvidia traded around 65 times next year’s earnings. A year ago, it was about 35.
Today, Wall Street expects its earnings per share to go from about $9.31 this fiscal year to about $15.91 next year. The stock is up only about 20% this year, so at around $220, it trades for about 14 times next year’s earnings.
The S&P 500 trades at 19 times. So the one company that will almost certainly still dominate AI 3 years from now has lost most of its multiple while its business grew several times over.
Economists call that a higher risk premium, the extra return investors demand for owning stocks over safer assets. When investors want more return for the risk, they pay less today for profits that are years away.
AI companies get hit hardest, because so much of their value sits in those far-off profits. The whole group is going through this, Micron and Marvell included, and the only way out is earnings growth big enough to outrun the shrinking multiple.
Why The Selling Looks So Ugly
Hedge funds had a record first half of the year, mostly riding the AI trade. Since early June, it’s been one long digestion.
The risk models they use lean on how stocks and bonds have moved together in the past. Those relationships are breaking one by one, so the funds have to cut their borrowing, and they cut it where they made the most money.
That’s why it feels like a crash even while demand for AI keeps growing. Too much leverage chasing too few trades is a positioning problem, and positioning problems resolve.
Where The Money Goes Next
In the next 3 to 5 years, the money goes to companies whose earnings jump far enough to beat the shrinking multiple. That means memory, where revenue has gone up 11-fold, the power and cooling companies solving the heat problem, and the custom chip and networking companies with contracted orders.
To know when the AI trade is over, watch what it costs to rent the chips and what it costs to buy memory.
As long as both keep rising, demand is still outrunning supply, and a selloff is an air pocket, a chance to buy. When they turn lower, it’s time to get out.
The rule for this market is simple. Own the businesses whose earnings grow faster than their multiples shrink, buy the air pockets, and never sell a stock because its multiple fell.
Tonight Is The Next Test
Meta reports after the close, with its call at 4:30 p.m. ET. It’s one of the two biggest drivers of earnings growth for the Magnificent 7 this quarter, along with Nvidia.
If Meta’s numbers are strong and the stock still gets hit, the multiple is shrinking again, and this rule says to buy that kind of air pocket.
Licia Leslie is going live this afternoon at 4:15 p.m. ET, and Meta and Tesla both report while she’s on. She’ll show you how to draw trendlines and trade the gaps that earnings leave behind, as the numbers hit.
It’s day 2 of her free 3-day series, and you can jump in today even if you missed yesterday.
===>Watch Meta’s Report Live With Licia
Your only option,
Mark Sebastian
P.S. Licia goes live at 4:15 p.m. ET, 15 minutes before Meta’s call starts. If tonight’s numbers create an air pocket, you’ll see it form in real time with her. Join Licia today here.