We Said the AI Rally Had Room to Run. Does This Chip Selloff Change Our Mind?

This is an opinion piece. It reflects our view and is not financial advice or a recommendation to buy or sell any security.

Back in February, we published a piece arguing that the AI rally had more room to run. The fundamentals, we said, looked stronger than the "it's all a bubble" crowd was giving them credit for. This week, chip stocks fell out of bed — a sharp, global selloff in semiconductors, led by memory names, that dragged the whole market down with it. So it's only fair that we do the thing too few commentators ever do: come back and grade our own homework.

What we still think we got right

The core of our February argument was that, unlike the dot-com era, today's AI leaders are real businesses with real revenue, real profits, and customers writing real cheques. That hasn't changed. The companies at the center of this build-out are still generating enormous cash flows, and demand for AI computing power is still, by every available measure, growing rather than shrinking.

The selloff this week wasn't triggered by AI demand collapsing. It was triggered by worries about supply, valuation, and crowding — concerns about whether too much capacity is being built and whether the stocks had simply run too far, too fast. Those are very different fears from "the technology doesn't work" or "nobody wants this." The underlying engine of the rally — actual AI adoption — is still running.

What we underweighted

Here's where we'll be honest about our blind spot. In February, we focused heavily on fundamentals and didn't say nearly enough about valuation and concentration risk.

A great business can still be a dangerous stock if its price has been bid up to assume perfection. When an entire sector is priced for a flawless future, it doesn't take a disaster to cause a selloff — it just takes "slightly less perfect than expected." That fragility is exactly what we underplayed. We were arguing about whether the AI story was real (it is) when the more important near-term question was whether the prices had gotten ahead of even a real story (they had).

We also underweighted how dangerous it is that so much of the market now rides on a handful of names. As we wrote this week in concentration risk, explained, when a few giant stocks dominate the index, their bad days become everyone's bad days — including index-fund holders who thought they were diversified. A narrow rally is a fragile rally, and we should have stressed that more.

So — have we changed our mind?

Partly. Here's our honest, updated position:

• On the technology: No change. We still believe AI is a genuine, durable shift, not a fad, and that the leading companies are real. A multi-year theme can absolutely survive a brutal few weeks for its stocks.
• On the stocks at these prices: More cautious than we were in February. "The story is real" and "the stocks are a good buy today" are not the same sentence, and we conflated them too easily before. When valuations are stretched, the range of possible outcomes widens in both directions.
• On what investors should do: Exactly what we always say. Don't try to call the top or the bottom of something this volatile. If you're broadly diversified and investing on a schedule, a chip selloff is just more shares on sale; if you're concentrated in a few AI names you bought at the peak, this is the lesson about why concentration cuts both ways.

The bigger lesson

We'd rather publish this awkward reassessment than quietly memory-hole our February call. Markets humble everyone who makes predictions, and the only thing worse than being wrong is pretending you weren't. The truth is somewhere in the messy middle: the AI thesis is intact, the valuations got frothy, and the stocks are now reminding everyone that "real" and "cheap" are different words.

If you want the deeper dive on where the chip sector actually stands, we broke it down in AI & Semiconductor Stocks in 2026. And if you're tempted to "buy the dip" here, read our take on when that actually works and when it doesn't first.