Remember that scene in The Big Short where Christian Bale’s character sits alone in a quiet office, drumming on his desk while the entire world parties on subprime mortgages?
That’s exactly what I picture when I think about Michael Burry right now.
While the rest of Wall Street throws billions at anything with “AI” in the pitch deck, Burry has gone suspiciously quiet—he even deregistered his hedge fund so he no longer has to show his cards every quarter. But make no mistake: the man who saw the housing bubble when everyone else saw endless growth hasn’t retired. He’s just loading up for the next one.
And this time, the target is artificial intelligence.
The Man Who Called the Last Crash Is Betting Against the Future
Most people know Burry from the movie or the book. Fewer remember that after the financial crisis, he largely disappeared from the public eye, made a fortune on water rights and farmland, then came roaring back betting against Cathie Wood-style growth stocks in 2021.
Now he’s done talking to the SEC altogether. In investment circles, that usually means one of two things: either you’re winding down… or you’re building a position so large you don’t want anyone front-running you.
Everything we know points to the second option.
The Quiet Genius Behind the Thesis
Before he left to start his own firm, Burry’s right-hand analyst was a guy named Phil Clifton. In his goodbye letter to investors, Burry didn’t hold back—he called Clifton the smartest mind he’d ever worked with. That’s not small praise coming from a medical-doctor-turned-investor who taught himself mortgage-backed securities from scratch.
Clifton’s research memos are now circulating among a small group of hedge fund managers, and let’s just say they read like autopsy reports on the AI boom.
“Just because a technology is good for society or revolutionizes the world doesn’t mean that it’s a good business proposition.”
– Phil Clifton, former Scion analyst
That single sentence sums up the entire bear case.
The Numbers Don’t Lie—They Scream
Everyone loves to quote the explosive growth stats: ChatGPT reaching 100 million users faster than any app in history, more than 60% of Americans now using AI tools weekly, enterprises racing to “not get left behind.”
But here’s the part nobody wants to talk about at parties.
The entire generative AI industry—OpenAI included—is on track for maybe $25-30 billion in total revenue this year. Maybe.
Meanwhile, the big cloud providers have quadrupled their capital spending to nearly $400 billion annually, with plans to drop another $3 trillion over the next five years building AI data centers.
Do that math in your head for a second.
Three trillion dollars of concrete, copper, and GPUs to support an industry currently making pocket change in comparison. Even if revenue grows 100x in five years—which would be the fastest scaling in business history—you’re still looking at infrastructure spending that dwarfs actual cash generation by an order of magnitude.
The Ghost of Telecom Past
Burry and Clifton keep pointing to one historical analogy that makes my skin crawl: the fiber-optic bubble of 1999-2001.
Back then, telecom companies spent hundreds of billions laying fiber across the planet because “internet traffic was doubling every three months.” Sound familiar?
- Capacity utilization in the U.S. eventually fell to about 5%
- Wholesale bandwidth prices collapsed 70% in a single year
- Companies like Global Crossing and WorldCom went bankrupt despite having the “best” technology
Clifton’s argument is brutally simple: we’re doing the exact same thing, just with GPUs instead of fiber.
Microsoft, Amazon, Google, and Meta are in an arms race to build the most data centers, signing power contracts that would embarrass small countries, all based on the assumption that demand will eventually catch up.
But what happens if it doesn’t?
Already we’re seeing cracks. Microsoft quietly canceled 2 gigawatts of data center projects earlier this year. Alibaba’s chairman went on Chinese television and warned—on the record—that an AI infrastructure bubble is forming.
Nvidia: The Pickaxe Seller in a Gold Rush That Might Not Have Gold
No company embodies the AI frenzy more than Nvidia. The stock is up more than 2,000% since late 2022. Jensen Huang can’t build chips fast enough.
But here’s where the Scion research gets vicious.
The cloud giants are depreciating their servers over six years on the books. Nvidia, meanwhile, releases a new architecture every single year that makes the previous generation look like a Commodore 64.
So which is it?
Either the new chips aren’t that much better (in which case why are customers panic-buying?), or the old chips become economically obsolete way before the accountants finish writing them off.
Burry’s team thinks it’s the second one. And when companies finally admit that billions of dollars of “assets” are actually doorstops, the write-downs are going to be biblical.
Nvidia can’t have it both ways: claiming revolutionary leaps with each generation while insisting the old hardware stays valuable for years.
I’ve spoken to data center engineers off the record. They laugh at the idea that a three-year-old GPU cluster stays competitive when power efficiency is everything. The moment a new Hopper or Blackwell chip ships, the old ones get relegated to training tiny models—if they’re lucky.
The Demand Side Looks Even Worse
Let’s be generous and say AI adoption explodes. Let’s say every company on Earth starts using generative AI for 10% of knowledge work by 2030.
Even then, the economics are brutal.
Running a single high-end query today costs real money in electricity and compute. Most companies I talk to are already hitting the “cool, but not $50 million cool” wall when they run pilots.
Consumers? They’ll use ChatGPT until the free credits run out, then go back to Google search. We’ve seen this movie before with 3D TVs, VR headsets, and the metaverse.
Amazing demos. Lukewarm wallets.
Why This Time Feels Different (And Why It Isn’t)
Every bubble has its defenders who say “this time is different.”
They’re usually right—until they’re catastrophically wrong.
AI really is transformative. I use it every day. It writes better code than half the junior developers I’ve hired, summarizes research papers in seconds, and makes my dumb ideas sound almost intelligent.
But transformative technology and profitable investment are two completely separate things.
Railroads changed the world. Most railroad companies went bankrupt.
The automobile created trillions in value. Hundreds of car companies from the 1920s no longer exist.
The internet is the single greatest wealth-creation event in history. Remember Pets.com?
What Happens If Burry Is Right?
In a sane world, the correction would be orderly: capex gets cut, valuations come back to earth, Nvidia trades at 30x earnings instead of 70x, and we all move on.
But markets rarely do sane.
When the narrative flips from “infinite growth” to “what were we thinking,” the selling can be violent. Especially when passive index funds and retail options traders are leveraged to the hilt.
I’m not predicting 2008-level carnage. But a 50-70% drawdown in the magnificent seven stocks? That’s well within historical precedent for this kind of mania.
Final Thoughts: Respect the Asymmetry
Here’s what keeps me up at night.
Being early on a bubble call is painful. Burry was laughed at for years before the housing market cracked.
But being wrong on the upside costs you maybe 30-50% if you sit out.
Being wrong on the downside when you’re fully invested? That can wipe you out for a decade.
Michael Burry doesn’t make bets unless the asymmetry is ridiculous in his favor.
The fact that he’s willing to stake his reputation—again—on artificial intelligence being the next big mispricing should at least make you pause before you YOLO your rent money into the latest AI ETF.
History doesn’t repeat itself, but it sure rhymes.
And right now, the rhyme sounds an awful lot like 1999.
