Michael Burry, the investor who became a household name for predicting the 2008 mortgage collapse, is placing fresh money on the bet that the AI boom will end badly, or at least badly enough to hammer Nvidia's stock by more than 40 percent before early 2027.
In April 2026, Burry added to his bearish position by purchasing January 2027 put options on Nvidia at a $115 strike price, paying a premium of roughly $3.30 per contract. At the time, Nvidia was trading near $188. He is also still holding earlier January 2027 puts at a $100 strike. The math is stark: those bets only pay off in a big way if Nvidia's share price craters by more than 40 percent within the next several months.
The trade is not a wild swing. Burry has described it as roughly 3 percent of notional value, a controlled wager, not a portfolio-defining all-in. But the direction of the bet, and the man making it, should concentrate the mind of anyone watching the largest capital deployment cycle in American corporate history.
As of May 2026, the combined capital spending of Amazon, Google's parent Alphabet, Microsoft, and Meta is on track for $700 to $725 billion this year alone. That figure covers AI data centers, high-end chips, power infrastructure, and related hardware. These four companies are in what can only be described as full sprint mode, each pouring hundreds of billions into facilities and equipment because none wants to fall behind the others.
Nvidia sits at the center of the rush. It supplies the high-end GPUs that every major AI player needs, a position that draws comparisons to Cisco during the late-1990s internet buildout, the "picks-and-shovels king" that armed both sides of the networking boom.
That comparison should give investors pause. Cisco's stock peaked during the dot-com era and never recovered its highs. The telecom boom of the late 1990s and early 2000s saw companies race to lay fiber-optic cables and build networks. Overcapacity followed. Then came bankruptcies, an 80-plus-percent crash in telecom stocks, and massive write-downs. The infrastructure survived and eventually proved useful, but the shareholders who funded the buildout got wiped out along the way.
The surge in AI-related imports already visible in trade data shows just how much capital is flowing out the door to build this new layer of computing power.
Burry's thesis rests on a pattern that has repeated across centuries of American and British economic history. The technology is real. The demand is real. And the losses are still real.
Railroad mania swept Britain in the 1840s and the United States through the 19th century. Thousands of miles of track were laid. The railroads transformed commerce and connected the continent. But the speculative frenzy that financed the buildout left investors holding worthless shares when too many lines chased too few passengers and too little freight.
Before the railroads, states and private firms in the early 1800s rushed to dig canals after early successes like the Erie Canal, which boosted New York's dominance. The canal boom ended in the Panic of 1837, which brought defaults and pain across the country.
In each case, the underlying technology changed the world. The investors who funded the overcapacity phase paid the price. The question Burry is posing with real money is whether AI infrastructure spending has entered that same dangerous zone, where every major player builds simultaneously, driven by fear of falling behind, and the result is a glut that crushes margins and valuations.
The concern is not theoretical. Communities across rural America are already pushing back against the physical footprint of the data center expansion, a sign that the buildout has moved well past the planning stage and into the landscape itself.
Some market skeptics have gone so far as to call the current AI spending cycle "the greatest capital misallocation in history." That label may prove too dramatic, or it may prove too mild. But the worry is spreading beyond contrarian hedge fund managers.
Breitbart reported that Burry's Scion Asset Management had previously disclosed put options on both Nvidia and Palantir, signaling a bearish view tied directly to bubble concerns. Investor worries have also grown over what some describe as circular deals involving OpenAI, Nvidia, and other AI firms, arrangements that raise questions about whether parts of the boom are being artificially supported by money cycling between related parties.
Jeff Bezos, the founder of Amazon, one of the very companies spending the most on AI, has acknowledged the dynamic plainly. He described AI as part of an "industrial bubble," saying both good and bad ideas get funded during periods of market excitement. "The good ideas and the bad ideas. And investors have a hard time in the middle of this excitement, distinguishing between the good ideas and the bad ideas," Bezos said.
When the man writing the checks admits that distinguishing good bets from bad ones is nearly impossible in the middle of the frenzy, the rest of us should pay attention.
Demand for AI compute is real and growing fast. Early wins in coding assistance, search, and productivity tools have given corporate buyers enough confidence to keep signing contracts. The technology is not vaporware. Companies are deploying it, customers are using it, and measurable gains exist in specific applications.
That is precisely what makes this cycle dangerous. The dot-com bust did not happen because the internet was fake. It happened because real demand was used to justify fantasy-level capital deployment. The canals worked. The railroads worked. The fiber-optic cables worked. The investors still got crushed.
The race to build specialized AI chips at companies like Google shows how deeply the infrastructure competition has penetrated, firms are not just buying Nvidia's hardware but designing their own, a sign of both genuine demand and the kind of arms-race logic that can lead to overcapacity.
For his puts to pay off in a meaningful way, Nvidia's stock would need to fall from roughly $188 to below $115, a decline of more than 40 percent, by early 2027. That is not a modest correction. It implies something has gone seriously wrong: underutilized data centers, a sharp capex cool-down from the hyperscalers, hardware obsolescence as new chip architectures emerge, or a broader market repricing of AI valuations.
Burry says he expects some clarity on his thesis by late this year. That timeline suggests he is watching for signs that Big Tech's spending commitments are softening, that utilization rates at new data centers are disappointing, or that the revenue growth AI was supposed to generate is not materializing fast enough to justify the capital outlays.
The broader economic effects of this spending binge are already rippling outward. Analysts have flagged AI-driven cost pressures showing up in inflation data and household expenses, a reminder that the buildout's consequences extend well beyond stock tickers.
The conservative instinct here is straightforward: when everyone in a room is making the same bet at the same time for the same reason, fear of being left behind, the odds of a painful correction rise sharply. That does not mean AI is a fraud. It means that $700 billion in annual spending by four companies, all chasing the same customers with the same technology, carries concentration risk that markets have historically punished.
Nvidia's position as the dominant chip supplier magnifies the stakes. If spending slows, Nvidia feels it first and hardest. If a new architecture renders current GPUs less valuable, Nvidia's premium pricing evaporates. If even one or two hyperscalers pull back, the ripple hits Nvidia's revenue line before it hits anyone else's.
Michael Burry has been wrong before. He was early on the housing trade by nearly two years, and early in markets often looks identical to wrong. But his track record earns him a hearing, and the historical parallels he is drawing, canals, railroads, telecom, are not abstract. They are the actual record of what happens when real technology meets unlimited capital and unlimited fear of missing out.
When the smartest money in the room starts betting against the hottest trade on Wall Street, the rest of us would do well to at least ask the question the cheerleaders refuse to: what happens when $725 billion worth of data centers need customers who haven't shown up yet?