The AI Selloff Is Not the End. It Is the Setup.
Listen to the audio version of this article (generated by AI).
In 1908, the American automobile business looked like it was eating itself alive.
More than 240 companies were building cars that year, and dealers could not tell one from the next. Money poured in, then vanished in a season. Buicks and Oldsmobiles piled up in showrooms nobody could afford to rent. The panic of 1907 had just rattled every bank in the country, and the newspapers were calling the automobile a fad for rich men that regular families would never buy in bulk. Investors who had ridden the boom for two or three good years watched their shares get cut in half and started asking whether the whole thing was a mirage.
Within two decades, the automobile industry built the middle class, paved the country, and created some of the largest fortunes in American history. The demand never stopped growing. What stopped, again and again, was the market’s patience for waiting to see it confirmed.
I am telling you this because I think you are experiencing the same thing right now in AI infrastructure stocks, and I think most investors are about to make the same mistake those 1908 skeptics made.
Here is my promise to you: This sell-off in AI infrastructure stocks is a gift, and I believe the stocks that dropped 20% to 30% over the past few weeks could roar back 50% to 60% within a matter of weeks.
But the problem keeping you from buying is the fear that AI chip demand is about to flatten out just as a wave of new supply hits the market, which is the exact setup that has ended semiconductor booms before.
That fear, however, is overblown: Samsung Electronics Co. reported preliminary second-quarter operating income of nearly 90 trillion won, up 19 times year over year and well above expectations, describing memory demand as white-hot, and the market sold the stock anyway.
Everyone is watching the wrong data. And I will walk you through exactly what to watch next, and which smaller, high-torque semiconductor names I recommend buying into this dip, because the real answer arrives in three weeks, not today. See more in our latest episode of Being Exponential With Luke Lango:
Why the Market Is Ignoring Great News
Samsung’s blowout quarter should have sent AI infrastructure stocks higher. Instead, the VanEck Semiconductor ETF (SMH) dropped 4% to 5% that same day, and the Nasdaq fell more than 1%. Micron Technology Inc. (MU) and Western Digital’s (WDC) SanDisk (SNDK) spin-off both fell 10% the following day, right on the heels of Samsung confirming that memory demand is roaring in the present tense.
The market does not care what is happening right now. It cares what happens over the next six to 12 months, because a wave of new chip supply is finally arriving, and the entire bull case for this trade rests on demand staying ahead of it.
That tension, supply catching up to demand, is what is dragging these stocks lower. It has nothing to do with whether the AI Boom is real.
The Buy Zone, By the Numbers
The SMH ETF sits about 14% below its highs, and history says that is exactly where AI semiconductor stocks tend to bottom. Every routine pullback in this group since the AI Boom began in late 2022 has bottomed in the 10% to 15% range, with two exceptions: a 24% drop in late summer 2024, and a 35% drop during the Liberation Day tariff scare. Outside those two, this marks the thirteenth double-digit correction in AI infrastructure stocks since the boom started, and semiconductor stocks are up more than 565% since 2023 despite living through all thirteen.
Zoom out further, and the pattern holds for the entire history of technology bull markets. Semiconductor stocks endured multiple 10% corrections and two nearly 40% bear markets between 1995 and 2000, then still climbed more than 1,100% from the start of that stretch to the March 2000 peak. Sharp pullbacks are the price of admission for triple-digit rallies. You do not get one without the other.
The Fundamental Case Has Not Moved
Here is what actually matters, and it is not the war in Iran, not the Federal Reserve, and not oil prices, unless those things get extreme enough to threaten one specific decision: how much Amazon.com Inc. (AMZN), Meta Platforms Inc. (META), Alphabet Inc. (GOOGL), Oracle Corp. (ORCL), and Microsoft Corp. (MSFT) plan to spend on AI infrastructure.
Those five companies are spending roughly $800 billion this year on AI infrastructure, a figure that could top $1 trillion annually by 2027. Amazon just moved to raise another $25 billion in bonds specifically to fund AI infrastructure, pushing total AI-related debt issuance toward $335 billion this year, more than double last year’s pace. Cash-rich companies are tapping debt markets because they have already emptied their own coffers, and that behavior signals acceleration, not retreat.
Meanwhile, OpenAI recently raised $122 billion, Anthropic raised roughly $60 billion, and SpaceX Inc. (SPCX) raised $85 billion and is emerging as a serious compute player in its own right. None of that points toward these companies pulling back on capital expenditures when they report earnings in roughly three weeks. I expect they reaffirm, and likely raise, their 2026 capex guidance, with bullish directional commentary on 2027 and 2028 spending plans.
Where I Am Putting Money to Work
I recommend running a screener for stocks up sharply over the past six to 12 months, down more than 10% from their 52-week highs, yet still trading above their 200-day moving averages, meaning the long-term uptrend stays intact even during the pullback.
Four names fit that profile in the semiconductor equipment and materials space right now. I recommend Aehr Test Systems (AEHR), a core part of the AI supply chain currently pulling back within its longer uptrend. I recommend Ultra Clean Holdings Inc. (UCT), whose cleaning and inspection equipment is mission-critical to chip manufacturing and whose chart still looks constructive. I recommend Axcelis Technologies Inc. (ACLS) and ACM Research Inc. (ACMR) as two more names positioned to move higher once technical support confirms.
The Verdict
Volatility is a feature of every hypergrowth bull market I have ever traded. The 1900s automobile shakeout felt like proof the industry was collapsing to the people living through it. It was proof the industry was working exactly as growth industries do: in violent, uneven bursts that reward patience and punish panic. I see the same pattern in AI infrastructure stocks today, and I believe late July, when the hyperscalers report earnings and confirm their capex plans, is the moment this trade reawakens.
Want the full breakdown, charts and all? Watch this week’s episode of Being Exponential, and drop your questions in the comments for a future show, or send them here.