By 2027, Nvidia Or A Cloud Giant Will Suffer A 25% Slide

My call: before New Year’s Eve 2027, at least one of the following will be down 25% or more from a local high on closing prices: Nvidia or one of Amazon, Microsoft, Alphabet, or Meta.

If that sounds bearish, look at what has to go right to avoid it. The world’s most cyclical industry has to behave like a utility, the largest capex binge in tech history has to be perfectly timed, and every regulator and central banker on earth has to act like an AI influencer.

That is the consensus trade. Michael Burry, naturally, is on the other side. He has walked away from simple shorts and loaded up on long-dated Nvidia puts in the mid-$100s that expire in September 2027. In plain English, he is paying real money for the right to profit if the market’s $5 trillion AI mascot loses more than half its value.

The bet: one brutal drawdown by 2027

Let us define the scorecard. Between now and December 31, 2027, I expect at least one of these names to log a 25% or bigger peak-to-trough decline on closing prices: Nvidia, Amazon, Microsoft, Alphabet, or Meta. No intraday flash crashes, no technicalities. A visible air pocket you can circle on a price chart.

This is not a call that AI is fake or that GPUs go the way of Beanie Babies. It is a call that valuations, margins, and capex expectations do not glide gently forever. Structural winners still hit potholes large enough to spill your drink.

Driver 1: a once-in-a-generation capex party

The bull script is clean. Nvidia is selling every data center GPU it can fab. Gross margins hover near 75%. EPS rose north of 120% year over year. Management just added a record $150 billion to its buyback authorization, bringing the total up to $235 billion through fiscal 2028. Wall Street looks at that, looks at a mid-20s forward P/E, and declares the stock “cheap.” Some analysts talk about a double by 2027 and price targets north of $350.

Behind that confidence is the hyperscaler firehose. A few giant cloud customers account for tens of billions of Nvidia’s data center revenue. Those same giants are telling investors AI capex will run hot for years, and second-derivative plays like Micron and Super Micro are waving multiyear order books into 2027. HPE is rolling out racks full of AMD GPUs as if the power grid were an optional side quest.

This is the sort of synchronized buildout that creates religions on the way up and inquiries on the way down. Everyone is extrapolating today’s shortage into a permanent condition. No one is running investor decks titled “Here Is The Part Where We Accidentally Overbuild $300 Billion Of Compute.”

Driver 2: cyclicality does not care about your TAM slide

Semiconductors are not new. Neither is disappointment. Nvidia’s own history is a reminder: in fiscal 2023, right before the current AI vertical takeoff, revenue was flat and net income fell 55%. That was before the company became the de facto monetary policy of AI.

Every big compute wave has followed the same script. PCs, smartphones, crypto, cloud: breathless growth, tight supply, then a period where demand is merely “very good” instead of absurd. Earnings stop surprising. Multiples compress from story-stock to normal-stock. The heroes of the last cycle are still heroes, just 30 to 50% cheaper for a while.

This time the cycle is concentrated in fewer hands. Hyperscalers account for a majority of Nvidia’s data center haul. If half a dozen people can move your revenue line by tapping the brakes, you are not diversified, you are an interest group.

All it takes for a 25% drawdown is one of three very ordinary things: a plateau in GPU orders, visible migration to in-house or AMD accelerators, or an investor realization that AI revenue is not catching up with AI capex on the promised timetable. None of those require an AI “bust.” They just require grown-up budgeting.

Driver 3: buybacks and puts on the same timeline

The most interesting part of this story is that both sides, the faithful and the short, are circling the same date.

Nvidia’s historic buyback program runs through fiscal 2028, which aligns neatly with bulls who assume today’s earnings power is a floor that stretches at least that far. One optimistic model has fiscal 2027 EPS around $9 and argues that 25 times earnings gets you near today’s price, which is pitched as reassuring. The subtext: we can stay here or higher if earnings simply do not collapse.

Burry read an Ares Management report on AI capex and did the opposite. He pulled his AI bubble timeline forward from 2028 to 2027, then published that shift in his Cassandra Unchained newsletter. His Nvidia puts, struck in the mid-$100s and expiring September 2027, start to sing if the stock falls more than half from roughly $230.

Both sides are telling you the same thing: the real test of this entire AI infrastructure regime is the 2026-2027 window. Either earnings and AI revenue step into the footprint that current capex is paving, or someone has to explain to shareholders why they built a second internet to sell the same software subscriptions.

Stakes: who actually eats the hangover?

Because this is the Prediction Desk, let us be precise. This call resolves true if, between publication and December 31, 2027, at least one of Nvidia, Amazon, Microsoft, Alphabet, or Meta closes a day 25% or more below any previous closing high in that window.

My base case is that Nvidia is the softest target. It sits directly in the blast radius of any AI capex rethink, it trades on margins that leave very little room for regression, and it is being defended by the loudest pile of buybacks in history. That is a familiar setup: companies that repurchase most aggressively at peak margins often end up funding their own downside volatility.

The cloud giants are not safe, only diversified. Their AI narratives are underwriting huge budget lines, yet the monetization evidence is mostly vibes and “run rate” numbers with creative rounding. If AI revenue does not show up in force by the mid-2020s, the market will not keep rewarding capex as performance art.

For policy people and allocators, this is the risk that matters. Governments and corporates are anchoring multi-year AI spending plans to a market price signal that currently assumes no major air pockets. If the signal turns out to be a sentiment bubble, the payback arrives as budget cuts, slower buildouts, and a fierce search for someone else to blame.

Verdict: gravity will get at least one of them

I am not betting on Burry’s exact strike. Sub-$100 Nvidia by September 2027 requires a proper crash. The forecast here is simpler and, in my view, higher probability: at least one of Nvidia or the Big Four cloud giants will trade a quarter off its highs before 2027 is out.

To be wrong, we need four smooth years of record AI capex, flawless earnings execution, friendly rates, contained regulation, and an investor base that never once panics about paying peak multiples for peak margins. In other words, we need the most important technology shift in decades to behave like a Treasury bill.

Burry may or may not nail the size of the fall. On the existence of a fall, I am willing to plant a flag. When the dust settles, either I will have underestimated the market’s appetite for permanent miracles, or Nvidia will finally learn what it feels like when the buyers run out of greater fools and start asking for greater profits.

Until then, enjoy the party. Just remember that under the strobe lights and GPU benchmarks, you are still dancing on a very old, very creaky thing called the business cycle.