Planet Money‘s coverage of generative AI has been consistently facile and far too reluctant to challenge or examine critically the claims of AI proponents. Their most recent episode about this, “So are we in an AI bubble? Here are clues to look for,” was so bad that I felt compelled to send them this email about it:
To: planetmoney@npr.org
Subject: What you missed about the AI bubble
I just finished listening to https://www.npr.org/2026/01/09/nx-s1-5672643/market-ai-what-is-a-bubble. It was hugely problematic.
First of all, you failed to acknowledge or mention or even hint at in any way the fact that bubbles are very often driven by at best smoke and mirrors and at worst outright cheating and fraud.
I don’t have to tell you about all the fraud that caused the housing bubble, because you reported on it extensively and in great detail.
As for the dot com bubble, I was there, working for a company whose IPO paid for the down-payment for the house I still live in 29 years later, and I can tell you first-hand that all of us engineers working for that company knew that senior management was delusional and our IPO was driven by lies. Not to mention all the jokes that were being told during that bubble, of which you surely must be aware, about how all any startup had to do to to get a huge pile of investor cash was put the word “internet” in their pitch somewhere.
There is every reason to believe this same thing is going on with AI. There have been plenty of news articles explaining why—much of it straight economics reporting your team should surely be aware of—and plenty of rock-star experts in the field calling it out on a regular basis.
You failed to acknowledge any of this. Instead, you begged the question by assuming that the market efficiency hypothesis (i.e., Fama) is correct, and then the rest of what you said for the entire episode was predicated on that assumption. Markets where investment is being driven by lying, cheating, and fraud are not efficient, they are rigged. Rigged markets cannot be sustained forever; this is why bubbles pop.
Second, the bit at the end about “maybe all these data centers we’re building will be useful for something else even if the bubble pops” entirely missed the point—which, again, is no secret, it’s mentioned frequently in news coverage—that these data centers are being built specifically for AI with hardware which is only good for AI and which becomes obsolete and has to be replaced at gargantuan expense every few years. This is nothing like the fiber optics you mentioned as a supposedly comparable example. Fiber-optic cables can be used to carry any data for any purpose. AI data centers can’t be repurposed for other things without ripping out all the hardware inside them and replacing it, at huge financial and environmental cost.
There are plenty of experts who could have discussed both of these issues intelligently. It’s journalistic malpractice that you didn’t seek them out and include any of what they have to say for this episode.
Jonathan Kamens
