원문 보기
Twenty-six years ago, American companies poured money into fiber cables, servers and websites faster than at any time in modern history. Then the Nasdaq lost almost 80% of its value, and it took fifteen years to climb back. Now Michael Burry, the investor who saw the 2008 housing crash coming, says we are about to break that record.
His number is 2.07%. That's how much the biggest American companies are investing, after wear and tear, compared to the size of the entire economy. The last time it was this high, the dot-com bubble was bursting. So is AI next? Burry became famous for The Big Short. Before 2008, he bet against the American housing market while almost everyone else was buying.
Today he publishes his research on Substack, and last week he turned his attention to artificial intelligence. His metric is simple. Take every dollar the companies in the S&P 500 spend on new equipment, buildings and data centers. That's capital expenditure. Then subtract depreciation, the value their existing equipment loses every year as it wears out or goes obsolete.
What's left is net investment: how fast corporate America is really growing its pile of physical stuff. Divide that by the size of the whole US economy, and you get Burry's number. As of the end of June, it stood at 2.07% of GDP. According to Burry, that's higher than at any point in almost four decades, with one exception: the tech, media and telecom bubble of the late 1990s.
Why compare spending to GDP? Because it shows how big the bet is relative to everything else the country produces. A few billion dollars is noise. Two percent of the entire US economy, year after year, is a wave that everything else has to make room for. And the driver this time is obvious. Microsoft, Amazon, Alphabet, Meta and Oracle are building AI data centers as fast as they can pour concrete.
Burry expects the number to keep rising over the next few quarters, possibly past even the dot-com levels. Which raises a question. What happened the last time? Burry's point is that heavy investment doesn't stop when the stock market peaks. It keeps going. In March 2000, the Nasdaq hit its all-time high.
But companies kept spending for months afterward. The projects were already planned, the equipment already ordered. By October 2002, the Nasdaq had lost almost 80% of its value, and it didn't get back to its old peak until 2015. Fifteen years. Then came the hangover. All that equipment had to be written down.
Depreciation grew so large that for twelve straight quarters, from mid-2003 to mid-2006, net investment by S&P 500 companies was actually negative. Companies were wearing out assets faster than they replaced them. The same pattern showed up in the next cycle. The S&P 500, and the banks inside it, peaked in late 2007.
But investment kept going right through 2009, as the crisis took down Bear Stearns, Lehman Brothers, Washington Mutual and AIG. And in the 2010s, it happened again in America's shale boom. Energy companies drilled thousands of new wells. Energy stocks peaked in the middle of the drilling cycle, before the spending did, and then oil prices more than halved.
Three different decades, three different industries. And according to Burry, one common thread: when investment gets this large relative to the economy, it has ended in a bust. To be clear, that's a pattern, not a law of physics. Some booms leave behind something valuable. The fiber laid in the 1990s eventually carried the modern internet.
But many of the investors who paid for it never got their money back. So where's the risk hiding this time? Burry went through the financial statements of the five biggest cloud companies: Microsoft, Amazon, Alphabet, Meta and Oracle. And he found something that doesn't show up in the headline spending numbers.
Add up their future lease payments, their commitments to buy chips and hardware, their guarantees on other companies' debt, and the data centers still under construction. According to Burry, the total comes to about 3 trillion dollars. Reporting on his analysis breaks it down roughly like this: about 1.2 trillion in future leases and about 1.5 trillion in purchase commitments, with guarantees and other obligations making up the rest.
Most of that sits in the footnotes, not in the headline numbers. It's money these companies have promised to pay in the years ahead, whatever happens to AI demand. Think of a household that shows you its bank balance, but not its mortgage, its car loans, or the contracts it can't cancel. The balance looks fine.
The commitments tell you what the next ten years look like. At the same time, Burry says their free cash flow, the cash left after all this spending, is turning negative. And to keep building, they're borrowing more, and faster. If you watched our last video, you've seen this story up close. Oracle's quarterly capital spending more than tripled in a year.
Its free cash flow was negative 5.4 billion dollars last quarter. And it has 664 billion dollars in signed contracts it still has to deliver. There's one more problem, and it's about time. A data center takes three to five years to build, and the leases behind it often run thirteen to twenty years. But the AI chips inside it change every twelve to eighteen months.
Each new generation draws more power and needs different cooling. So a building designed for today's chips can be the wrong building for the chips of 2029. And the lease doesn't care. The payments keep coming. There's an accounting angle too. When a company builds a data center, it doesn't count the whole cost as an expense right away.
It spreads it out as depreciation over the years the equipment is supposed to last. If the equipment becomes obsolete faster than planned, what's left of its value has to be written off, all at once. That's why Burry thinks the reckoning won't arrive as a crash tomorrow, but as write-offs, perhaps in 2028 or 2029.
When commitments are this large, he argues, "a relatively small write-off has a bigger impact than we can now imagine." Now, the other side. Burry is famous for being right about 2008. He has also made plenty of calls that didn't play out on his timeline. And today's AI builders have something the dot-com startups didn't: enormous profits from their existing businesses.
Microsoft, Alphabet and Amazon can absorb losses that would have killed a startup in 1999. And demand for AI computing is real. Businesses are paying for it, and cloud revenue is growing fast. The question isn't whether AI is useful. It's whether the price being paid today can ever be earned back, and by whom.
Remember, in 2000 the internet wasn't a fad either. The internet won. Many of the companies that built it for everyone else didn't. But his core point is hard to argue with. Spending this large, this fast, has to earn a return. And for now, the spending is growing faster than the cash coming back. So is 2.07% a warning light, or the price of building the future?
Watch that number over the next few quarters. If Burry is right, it will keep climbing, past the dot-com record. And history says the hangover comes after the party, not during it. What do you think: is AI the new internet, or the next fiber glut? Tell us in the comments. If you want the money story behind the headlines, subscribe to Capital Letters Money.
Comments 0