Did the AI bubble just pop, or was that a dip to buy?

After the week we just had, everyone is asking the same question. Korea's market fell 10% in a session, the Nasdaq dropped more than 2%, chip stocks got torched, and the bounce that was supposed to fix it on Wednesday melted by lunch. So is this the moment the AI bubble finally pops, or is it just another scary dip that ends up being a gift? Here's my honest read, with both sides laid out, and where I come down.

The bear case: why this could be the top

Let's start with the scary version, because it's not crazy. Some of the most respected names in finance have been waving flags. Jamie Dimon and Ray Dalio both warned in the weeks before this selloff that something looked off. Michael Burry, the guy from "The Big Short," said back in May that the AI market reminded him of the final months of the dot-com bubble. One analyst put it bluntly: 2026 is starting to look a lot like 1999.

The valuation argument has teeth too. The S&P 500 has been trading around 23 times forward earnings, the most stretched it's been since the dot-com era. When stocks are priced that richly, you don't need bad news to fall. You just need the story to wobble, which is exactly what happened this week.

But here's the single best argument the bears have, and it's worth sitting with. The money going into AI is wildly out of proportion to the money coming out of it. An MIT study found that despite tens of billions poured into generative AI, about 95% of organizations are getting essentially zero return on it. Total AI revenue is running somewhere around $15 billion to $20 billion a year. To just break even on the infrastructure being built, that number needs to scale toward $160 billion a year. To actually generate good returns, you're talking $400 billion to $500 billion. That is an enormous gap, and it's the real thing that could end this trade.

The bull case: why this looks like a dip

Now the other side, and in my view it's the stronger one right now.

Start with the spending, because it's the opposite of what you'd see if the boom were ending. The four biggest spenders, Amazon, Microsoft, Google, and Meta, are on track to spend somewhere around $700 billion to $725 billion on capex this year, up roughly 77% from last year, with about three quarters of that going to AI. Amazon alone is near $200 billion, Microsoft around $190 billion, Google $175 billion to $185 billion, and Meta $115 billion to $135 billion. Bubbles don't usually burst while the buyers are increasing their orders by 77%.

Next, the valuations on the actual chip names aren't dot-com insane. The semiconductor index has been trading around 25 times forward earnings, below the 30-plus it hit at prior peaks. Micron, the stock everyone is fixated on this week, trades at roughly 7 times its 2027 earnings estimates while it's projected to grow earnings about 170% over the next two years. Even Nvidia, after a monster run, sits well below its own five-year average multiple on forward numbers. These are not the kind of prices you see at the top of a mania.

And the demand is real and visible. Micron has already said its entire 2026 HBM supply is sold out. Google recently placed an order with Intel for more than three million AI processors. When sentiment cracked earlier this month, it was a concrete order like that, not a feel-good speech, that turned it around. That tells you the buyers are still buying.

Finally, look at what actually drove this week's selloff. It wasn't a wave of canceled orders or a demand air pocket. It was a crowded trade unwinding, plus growing fear that the Fed will keep rates higher for longer, which makes all that debt-funded AI spending more expensive to carry. That's a real concern, but it's a rates-and-positioning story, not a "nobody wants the chips anymore" story. Those are very different things.

So, bubble or dip?

My take: this looks a lot more like a dip than a popped bubble, at least for now. The thing that actually defines a bubble bursting is demand evaporating, and demand isn't evaporating. It's growing. Capex is up 77%, supply is sold out, and the cheapest names in the group are cheap precisely because their earnings are growing faster than their stock prices. A selloff driven by interest rates and over-positioning is the kind that tends to get bought back, not the kind that wipes you out.

But I'm not going to pretend the bear case is empty, because it isn't. The ROI gap is real, and it's the one crack that could turn a dip into a burst. Right now the entire trade runs on faith that all this spending eventually produces the revenue to justify it. As long as the hyperscalers keep spending, the chip names hold up. The day those companies blink and cut their capex because the returns aren't showing up, that's when the bubble talk stops being talk. So the tell isn't really the chip stocks. It's the spending plans of the giants buying the chips.

What to watch from here

• Micron tonight. It reports after the close, and it's the cleanest near-term read on whether AI memory demand is still as hot as the bulls think. Strong guidance would go a long way toward ending this scare.
• Hyperscaler capex guidance. This is the real one. If Amazon, Microsoft, Google, or Meta start trimming their AI budgets, take it seriously. As long as they keep spending, the floor under chips holds.
• The Fed. A lot of this selloff is about rates, not AI. If the path turns less hawkish, a big chunk of the fear goes away on its own.

Bottom line

The bubble has not popped. What we got this week was a sharp, fear-driven dip in the most crowded trade on Wall Street, driven by interest rates and stretched positioning rather than a collapse in demand. The buyers are still buying, the chips are still sold out, and the leaders still look reasonably priced against their growth. That leans dip, not burst. The honest caveat is the ROI gap, the giant distance between what's being spent and what's being earned. Watch the hyperscalers, not the headlines. The moment they stop spending is the moment this story changes.

Sources

• Fortune, MIT report: 95% of generative AI pilots at companies are failing (MIT Project NANDA)

This is analysis and opinion for information only, not investment advice, and I'm not a financial advisor. Figures come from market reporting and analyst estimates as of June 2026 and vary by source, so do your own research and confirm the latest numbers before making any decisions.