Over the last three trading days, tech stocks, especially semiconductors and AI-adjacent names, have been crushed as investors rotated aggressively out of the sector. The Mag 7, the memory names and the broader semiconductor complex are all down sharply.

But this is not the first time. Since the AI boom kicked off in 2023, we have seen this story play out repeatedly. Each episode arrived with its own distinct headline scare, and each one, at least so far, was eventually bought.

Image Source: TradingView

Before getting into what's driving today's move, it's worth walking through the prior scares, because the pattern is instructive.

July–August 2024 — the monetization scare. This was the first real wobble. Google kicked off Big Tech earnings season with capital expenditures climbing sharply, and management struggled to give a clean answer on when that spending would translate into returns. Microsoft, for its part, framed AI monetization as something that would play out "over the next 15 years and beyond," not the near-term payback some investors were hoping for. The Nasdaq 100 fell more than 3% on July 24, its worst session since October 2022, and the anxiety bled into the violent early-August unwind of the yen carry trade. The core worry: the spending was unmistakably real, but the returns were not yet visible.

January 2025 — the DeepSeek shock. A Chinese startup claimed it had trained a model competitive with the leading US systems for under $6 million, using less advanced hardware. The read-through was that if frontier-level AI could be built far more cheaply, the case for hundreds of billions in GPU spending might be overstated. Nvidia lost roughly $593 billion in market value in a single session and the Philadelphia Semiconductor Index fell more than 9%, its steepest drop since the early-2020 COVID crash. Unlike the prior episode, this scare wasn't about slow returns, it was the fear that cheaper training would undercut the entire capex thesis. The selloff reversed quickly once the major hyperscalers reaffirmed their spending plans.

November 2025 — "AI bubble" fears. This one was a slower grind lower rather than a single-day crash, driven by stretched valuations and a growing chorus of skeptics. Michael Burry, of "The Big Short," argued that the hyperscalers were flattering their earnings by understating depreciation, extending the assumed useful life of AI chips and servers that, in his view, become obsolete far faster. The Nasdaq logged its worst week in months as institutional surveys showed a majority of investors believed AI stocks had become a bubble. The accounting angle made this a more sophisticated version of the bear case than earlier rounds.

Late January–February 2026 — the capex-guidance rout. This was the largest aggregate wipeout. A cluster of mega-caps shed well over $1 trillion in combined market value in a single week as fourth-quarter earnings revealed staggering capex plans, as Amazon alone guided to roughly $200 billion in infrastructure spending, a 56% jump and the highest commitment among the hyperscalers. A new fear joined the familiar one: not only was capex outrunning the cash flow funding it, but investors began to worry that AI itself was beginning to cannibalize the established software companies, the very names that had been considered safe AI winners.

An observation of my own: after years of watching markets, you'll notice these narratives often get assigned to the price action after the fact. Nothing in global markets happens in a vacuum, and prescribing a single tidy story to a selloff, while helpful for simplification, can be unhelpful for understanding the broader setup. Look closely and each of the four episodes above was triggered by a different worry, but each time an extended market that snaps its streak reaches for whichever AI-skeptic story best fits the tape that week.

Consider what else was happening underneath each "AI" selloff. The July–August 2024 drawdown is remembered as the "AI fatigue" trade, but the real violence came from the Bank of Japan's surprise rate hike unwinding the yen carry trade and a weak jobs report that tripped a recession indicator. The January 2025 DeepSeek shock hit a tape already on edge over a hawkish Fed, a 10-year yield near 4.7%, and fresh tariff threats. And the November 2025 "bubble" scare, while more genuinely valuation-driven, still rode on an unsettled Fed path and stretched positioning after a long summer melt-up. The AI story was the most quotable explanation each time, but it was hardly the only one.

That observation deserves its own deeper treatment, which we won't attempt here, but it's worth keeping in mind whenever a clean explanation gets attached to a messy move. Ultimately, markets move lower because there are more sellers than buyers, which is often an unsatisfactory explanation.

Other Factors Moving the Stock Market

In the current case, several variables are at play beyond the AI-spending headline.

Equities, tech and AI especially have been on a powerful run since the March lows that followed the onset of the US–Iran conflict. That rally pushed sentiment to heavily bullish extremes, and stretched positioning is precisely what leaves a market vulnerable to a sharp, fast reversal. When nearly everyone is already long and leaning the same way, there are few buyers left to absorb selling once it starts.

At the same time, interest rates have been grinding higher, pressured from two directions. Higher oil prices, a byproduct of the geopolitical backdrop have revived inflation concerns, while consistently robust labor market data has reduced the case for near-term rate cuts.  Together, those forces put upward pressure on yields and raise the prospect of a less accommodative Federal Reserve.

As far as I can tell, last Friday's strong employment report was the initial catalyst, by pushing rate expectations higher, while the renewed AI-spending fears added fuel to the fire. Overextended positioning then did the rest, leaving the tape vulnerable to a negative feedback loop of selling, which is what we are seeing today.

The Bear Case for AI Stocks Isn't Unreasonable

While I doubt this marks the end of the AI boom, it would be a mistake to dismiss the bears. They are raising legitimate points.

The sheer scale of the spending. More than $1 trillion has been poured into AI through data-center infrastructure and capital raised for the model labs — OpenAI, Anthropic, and others. For 2026 alone, the major hyperscalers have collectively guided to somewhere in the range of $600–700 billion in capital expenditures.

The opacity around returns. There is real uncertainty about the return on investment in these data centers, and the economics of actually running the models are murkier than they appear. When you pay a monthly subscription to an AI provider, the cost of serving your prompts may well exceed what you're paying, meaning the usage is being subsidized by an unknown amount. Loss-leading is not a new strategy, as several of the Mag 7 built their dominance by absorbing losses to capture markets first. But it has never been attempted at anything close to this scale, and the path to sustainable margins remains undefined.

The circularity. A growing concern is how interlinked the major players have become. Nvidia has invested in "neocloud" providers, companies that rent out GPU computing power, which in turn use that capital to buy more Nvidia chips. Nvidia has also committed to invest heavily in OpenAI, which has pledged to spend enormous sums on the very compute that flows back through the ecosystem. Supporters frame this as a "virtuous circle" that locks in scarce supply and critics see it as a web of interdependent commitments where a stumble at one node could cascade through the whole structure.

The coming mega-IPOs. A wave of richly valued, deeply unprofitable companies, the likes of SpaceX, Anthropic, and OpenAI are coming to public markets. By entering the major indexes while still burning cash on opaque business models, they could introduce fresh vulnerability for passive investors who hold them by default. Viewed cynically, the whole sequence can look like an opportunity for venture capital and other early backers to cash out at the top before any unraveling.

These are all reasonable concerns, but it's worth being precise about what they rest on: the assumption that data-center investment is structurally unprofitable. That is largely true today. It is far less clear that it will remain true. The margins on AI infrastructure are still being discovered, and history with prior technology buildouts suggests that early-stage unprofitability is not the same thing as permanent unprofitability. The honest position is that the verdict is genuinely unknown, which is exactly why the tape whipsaws on every new data point.

Which Stocks are Capturing the Flows

As money has come out of tech, it has been finding a home in the more defensive and beaten-down corners of the market. Today we're seeing real estate, consumer staples, healthcare, and some left-for-dead retail names catch a bid. Among the more interesting movers are Eli Lilly ((LLY Quick QuoteLLY - Free Report) ), Home Depot ((HD Quick QuoteHD - Free Report) ), Procter & Gamble ((PG Quick QuotePG - Free Report) ) and Starbucks ((SBUX Quick QuoteSBUX - Free Report) ), among many other established, cash-generative businesses that had been largely ignored while capital chased AI.

I'm not prepared to call this the start of a durable resurgence in these names, but the logic tracks. Most data points to a broadly healthy US economy even as investor attention has been monopolized by AI. If the economy continues to hold up, these unloved areas \could absorb a meaningful share of the flows rotating out of crowded tech.

This may raise more questions than it answers, and that's fine. Identifying the current environment is a more tractable task than predicting the future, and good portfolio management sometimes simply requires being appropriately defensive for the conditions in front of you today.