A short talk in lieu of a post. Back on full duty tomorrow.
Transcript:
Hi, Paul Krugman here.
I’m recording this on Tuesday afternoon. I just won’t have time to write a normal post for tomorrow when you’ll see this. And I would take the day off, except it seemed to me as if people might want some reaction to the carnage that’s been going on, at least in part of the tech sector and stock markets around the world, which has been pretty remarkable.
It’s really tempting to say that it’s deeply meaningful. But in general, you want to be very cautious about putting too much stake in stock market events. I’ll come back to that in a minute. But it is striking enough that it does seem to be worth commenting on.
So what’s happened? There’s been a fall in tech stocks very much concentrated in semiconductors. The Philadelphia Semiconductor Index was down almost 8%. on Tuesday. The KOSPI Korean Index, which is largely a semiconductor index, was down just about 10% sort of the previous day or the same day, you know, time zones. And there was a 2.2% fall in the NASDAQ. We’ve seen a lot of decline in tech stocks, things related above all to chips. What’s going on there?
Part of the answer is that trying to understand why the market does what it does is, generally speaking, a mug’s game. In this case, however, it does seem that part of what’s happening, probably a large part of what’s happening, is that the tone, the rhetoric surrounding use of AI, and hence the demand for compute, has really shifted quite a lot just very recently.
All of a sudden, we have a spate of studies that seem to show that, yeah, AI models allow people to churn out a lot more stuff, but the actual payoff to that stuff is much, much smaller than the volume of stuff that they’re churning out, most obviously lines of code, but just in general. AI lets you do much more, but how productive that is in terms of the ultimate goals of a business, let alone economic growth and quality of life is much more doubtful.
On top of that you have a rather abrupt, jarring turn in business strategy. Up until just the other day a lot of businesses were more or less whipping their workers into using AI — you know, we’re going to judge you on how much you’re using AI whether or not you really want to whether or not you yourself think it’s valuable. We’re actually going to score you, we’re going to require that you do tokenmaxxing.
And then, with compute getting scarce and with the price of chips having gone through the roof, suddenly the AI companies began charging and the marginal cost of using a lot of tokens became really, really very high. And suddenly companies were saying, oh wait, stop. We want you to economize on your use of tokens and hence to ultimately reduce the demand for compute. And that’s a sudden U-turn.
This is part of a broader phenomenon, which I’m going to write about very soon, which is that there is a kind of lack of organicness to the AI boom.
There are people who are using it because it looks great. They’re using it because it’s fun. I have colleagues who are just mucking around with Claude and finding some uses for it. But there’s also a large amount of Corporate America that thinks that this is the way it has to go. Fear of missing out, not by the individual investor, but by the corporate bureaucracy. And then pressure from the financial markets, saying, you know, your company better be on the cutting edge of AI or else. All of which is very fragile. It’s a kind of a bubble, but not in the normal sort of asset price form. It’s more of a kind of fad, almost a social delusion. And that, it seems likely, certainly got ahead of itself.
Now, I’m reading way too much into these stock prices. And so let me give you a little bit of a caution on all of that. So yeah, the Philadelphia Semiconductor Index was down 8% in a day, which is one hell of a drop. But it was up 157% over the past year.
So you want to have some perspective here. This is a stunning setback, but the fact of the matter is that over the course of a year, these stocks have been incredibly high-performing. The KOSPI, the Korean index, was down 10%, strictly speaking, 9.99%. But anyway, it was down 10%.
But after that 10% fall, it was up 172% over the year. So we’re not talking about a catastrophe. We’re not yet talking about, we aren’t even talking about a Bitcoin level of disappointment for investors. But okay, it’s a break in the trend.
The other thing we should say: the famous old line by my teacher and colleague, Paul Samuelson, was that the stock market had predicted nine of the last five recessions. There’s many more than that now. In fact, just over the course of the past year and a half, we’ve had two major stock market declines that turned out to be false alarms.
There was a big decline in April of 2025 after Liberation Day, the Trump tariffs, because there was a lot of people just sort of, it’s chaos, terrible things may happen. While the tariffs have been a bad thing, they did not cause an economic catastrophe and stocks recovered the losses that they experienced then.
And then there was another round of major stock declines associated with the Iran war. Of course, the Iran war has been a complete debacle and a disaster, and we’ll be paying a price for that for a very long time. But the consequences for short-run macroeconomics were more modest than many people, myself included, expected. And it appears that the Strait of Hormuz is going to gradually open because the United States basically said, okay, you win. It won’t literally say that, but in practice, that’s what we’re doing. So that is going to be over.
So it’s not that uncommon for the markets to react as if something terrible is about to happen and be wrong.
And so you really don’t want to assume — there’s a real temptation to assume — that because there’s so much money involved, a big decline in markets must be signaling that something is really very much amiss in the fundamentals, that where there’s smoke, there’s fire. And sometimes, no, there’s just smoke, no fire.
So this might not be that big a deal. But it comes at a moment when the rhetoric really has shifted. You can see that there’s just a kind of a walking back.
There was a really striking interview just the other day with Satya Nadella of Microsoft. Microsoft is actually a consumer of AI, rather than a producer. They have tools you can use within Microsoft products, but I think they run basically off OpenAI.
And Nadella was pretty scathing about saying, you know, we can’t give all of this power and all this money to the big AI companies, and we should be using cheaper models. And hinted that Microsoft may start making use of DeepSeek, the Chinese model, which is less comprehensive. In general, the Chinese models are less comprehensive, but immensely cheaper, and among other things, just do a lot less computation. That’s kind of the core of why they’re cheaper.
And in that case, the picture changes a lot.
What bearing does all of this have on AI and the future of the economy and AI and the future of humanity? Well, part of what we’re seeing may not be so much disappointment in what AI can do as realizing that this extremely compute-intensive AI is not essential.
And maybe you can still get whatever the big productivity benefits are and still possibly the big labor-displacing effects without quite so much compute. But it’s not entirely separate either. I think we need to be saying that this is what a quasi-bubble quasi-bursting might look like.
Take care.
