Mystery AI Hype Theater 3000

When the Bubble Bursts (with Matt Scherer), 2026.06.01

• Emily M. Bender and Alex Hanna • Episode 80

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 53:13

What should we do when the "AI" bubble pops? Emily and Alex are joined by researcher Matt Scherer, who recently published a report called "No Bailout for Big Tech Billionaires: Policies for when the AI bubble bursts." They unpack what the boosters are saying about their finances, and how we can all push back against arguments for a government bailout.

Matt Scherer is a fellow at the anti-monopoly think tank Open Markets Institute. He previously led the workers' rights project at the Center for Democracy & Technology.

References:

Fresh AI Hell:

Check out future streams on Twitch. Meanwhile, send us any AI Hell you see.

Find our book The AI Con here, and MAIHT3k merch here.

Subscribe to our newsletter via Buttondown.

Follow us!

Emily

Alex

Music by Toby Menon.
Artwork by Naomi Pleasure-Park. 
Production by Ozzy Llinas Goodman.

Alex Hanna: Welcome, everyone, to Mystery AI Hype Theater 3000, where we seek catharsis in this age of AI hype. We find the worst of it and pop it with the sharpest needles we can find. 

Emily M. Bender: Along the way, we learn to always read the footnotes, and each time we think we've reached peak AI hype, the summit of Bullshit Mountain, we discover there's worse to come. I'm Emily M. Bender, professor of linguistics at the University of Washington. 

Alex Hanna: And I'm Alex Hanna, director of research for the Distributed AI Research Institute. This is episode 80, which we're recording on June 1st of 2026. Our guest this week is Matt Scherer, who's a fellow at the anti-monopoly think tank Open Markets Institute. He previously led the Workers Rights Project at the Center for Democracy and Technology. Thanks for joining us, Matt. 

Matt Scherer: Thanks for having me. 

Emily M. Bender: We're really happy to have you with us. Matt has a report out last month called "No Bailout for Big Tech Billionaires: Policies for When the AI Bubble Bursts." You can find the link to that in the show notes.

Alex Hanna: So today, we're going to be taking advantage of Matt's economic expertise. He'll help us unpack what the AI boosters are saying about their finances and why we're not actually in an AI bubble. We'll also talk about how we can push back with arguments that we need a government bailout when the bubble bursts. 

Emily M. Bender: So here is our first delightful artifact. This is published in Forbes on May 26th of this year. The journalist's names is James Broughel, or Bruegel, we're not sure. And the headline is, "AI Can Change the World and Still Be a Bubble." 

Alex Hanna: Yes, and this is like an op-ed piece. It says he's a contributor, but it is written in a kind of an op-ed style. 

Matt Scherer: Yeah. I clicked through, and he is affiliated with the America First Policy Center. He used to be at the Competitive Enterprise Institute and the Mercatus Center. So this is not some left-wing techno pessimist dude that's just throwing rainwater on everybody's parade.

Emily M. Bender: Yeah. And that image, Alex, looks like something that we would've rejected for our book cover. 

Alex Hanna: Oh, yeah. It looks extruded from one of these things, but there is a bubble, and it just has a 3D lettering of AI in it, and then there is a sewing needle in it that's about to pop it. 

Emily M. Bender: Yeah, and the bubble has got this greasy sheen on it that, you're right, isn't quite right, which is probably what's suggesting this is extruded. So let's dive in. "Wall Street has spent the past year pricing in an artificial intelligence revolution. Now investors are starting to question whether the boom has gotten ahead of reality. Calling the AI boom a bubble may sound like one means artificial intelligence is fake or doomed to fail, but it actually means something far narrower. The price of AI-related assets may have climbed higher than the profits those assets can reasonably be expected to justify." I'm getting the intonation all wrong on this because his writing is weird. But anyway, "On that definition, the evidence now points to a bubble in AI investment." 

Matt Scherer: That's the understatement of the century, that "the evidence now points to a bubble in AI investment." At some point, if you scroll down like halfway in the article, there's a little graph that shows spending versus revenues on AI. There it is. So the far right, that's how much companies, enterprises, businesses are spending on AI products and services. The left bar is just how much essentially six tech companies are spending on AI this year. So I don't think that we've seen a gap of, what is that, $706 billion between what's being spent on a technology and what that technology is bringing in, ever in the history of economics. So there was a story in Wired this past year, and it described AI as the bubble to end all bubbles. If you could create the platonic ideal of a bubble, this is what it would look like. 

Emily M. Bender: Wait, so if we get through this one, we have the end of all bubbles and it doesn't happen again?

Matt Scherer: That's what I'm hoping to achieve in my fellowship. A bold ambition, but yes. 

Alex Hanna: Let's hope it doesn't happen again, because there's just no world left. It's like the old adage or, the old, I don't know what you'd call this, I don't know what weapons are gonna be used to fight World War III, but World War IV will be fought with sticks and stones.

Emily M. Bender: Yeah, or in other words, we're gonna get knocked back to tulip bubbles. 

Matt Scherer: Yeah, exactly. The next bubble will be in seashells. 

Alex Hanna: Yeah, overvaluation of really shiny rocks. So the next paragraph says, "The technology will still matter. Some people will get rich from AI. Consumers will get useful services. Scientists will get better research tools, and companies will find more efficient ways to produce goods and services. But none of that responds directly to the valuation question. A great invention can still be a bad investment if buyers pay too much for a claim on its future profits."

Emily M. Bender: So I don't disagree with that last sentence. And also, "some people will get rich from AI." That is also happening, because people are effectively selling this idea. But everything in between seems false. 

Matt Scherer: Yeah. I, like you, I sit on the extreme skeptical edge of where- or I guess, I would position myself on the extreme skeptical edge, I don't want to speak for you- on AI's potential impact on the economy and society. I have read "The AI Con," I suspect that y'all are where I am on that. But to me, it's interesting. I think that it is worth remembering he's writing for investors, basically. He's writing for people who are thinking about this from an investment business type perspective. And for that audience, this is a good way of framing it. Because that audience has been force-fed AI hype for the past two years. Everybody who is in any corner of corporate America right now cannot go 30 seconds without running into a, "This is a new AI tool that we want you to use," or, "This is a story about, hey, how AI is gonna transform your industry." He has to speak to that. And I do like the point that he is making of, even if you accept all that as true, that doesn't mean AI is not a bubble. And that is an important point for folks that are in that ecosystem to internalize. 

Emily M. Bender: Yeah, and we can agree with him there. I think I would say in terms of where I stand, I'm not on the extreme edge of anything. I am on very solid ground that this is all bullshit. There's no teetering on any edges out here. But yes, I share, and I don't, actually, I don't even like the framing of AI skeptic, because that says basically the question is, is AI a thing or not? And that's not, I think, the main question facing us right now. But is there a bubble or not is, I think, a good question to be looking at.

Alex Hanna: And I wanna bring something from the chat. sjaylett in the chat says, "And most of the people who will get rich are already rich." And magidin says, "Yeah, it should be some people are getting richer," end of paragraph.

Matt Scherer: They're gonna be a lot poorer in a few years, I suspect. But I suspect that they'll also still be multimillionaires and billionaires. 

Emily M. Bender: Yeah. Still at the top of the heap, but maybe without- yeah. All right, so this next bit is fun because it involves railroads. You wanna do the honors, Alex? 

Alex Hanna: Sure. So the subtitle, "A Revolution Can Still Be Overpriced," which is funny. "History is full of inventions that fit this pattern. Railroads transformed America, but many railroad investors lost money after the railroad bubble burst in the 1890s. The internet changed the nature of commerce, but most dot-com shares bought near the late 1990s peak were terrible investments. In other words, a technology can be useful, even transformative, while the stocks and private company valuations attached to it are too expensive."

Matt Scherer: That is the exact point that I make in a few things that I have in the ether. I've been writing so much on bubble-related stuff that I can't remember where I use the railroad metaphor and where I don't. But it's a great point. I would argue that railroads were actually the most transformative invention and development in transportation technology in history. It was the first technology that allowed people to travel long distances without using horses, essentially, which are much slower than railroads and obviously, there's all sorts of not just animal cruelty issues, but simple maintenance issues. A horse can only travel so far before it gets tired. Railroads did transform society. But when the railroad bubble burst in 1873, it caused the longest depression in the history of the United States, longer than the Great Depression. And that's despite the fact that railroad tracks last for a century before they wear down. So there was all this great infrastructure, that we are still in many cases using, actually, in some instances- either the actual track that was laid in the 19th century or stuff that's been repaired and replaced on it. But it still burst and caused immense economic damage. And again, one point that I'll make is that silicon chips and GPUs, those are not railroads. Those do not last a century. If railroads can be at the center of a bubble that causes immense economic damage when it bursts, then certainly AI, which is predominantly the money is being spent on this infrastructure with a much shorter shelf life, that can definitely do so. 

Alex Hanna: Yeah. And we'd also be remiss if we didn't mention just the immense labor exploitation and huge deaths of Chinese migrants coming to the US and then dying in the building of the railroads, and the ways in which that also, those communities were not benefiting from the growth. 

Matt Scherer: And the dispossession of Native Americans that tried to get in the way- or, not even tried to get in the way, that were already there and were forcibly removed from their lands in order to make way for them.

Alex Hanna: Yeah, and I think this railroad metaphor gets used quite a bit too, and I think one of those points about the ways in which GPUs and the hardware becomes obsolete rather quickly, is a point where that metaphor falls apart really quickly, especially because you were trying to think about the maintenance that will be required. It just needs to be continual. And the graph below, which we'll get to it in a bit, but the amount of investment that is being spent, so that's 725 billion. That was the original estimate that I think David Cahn, who was mentioned in this article, from Sequoia Capital mentions, is that it's 600 billion. And Cahn is effectively still saying, "It's still worth it. It's still gonna pay off." And just seeing the delta that hasn't shifted year over year, what kind of delusions do you have to be under that this gap can be shored up in any kind of meaningful way by enterprise sales? Or by really any sales, but enterprise is the big one. 

Emily M. Bender: Yeah, and Alex, you were making the point before we started recording that with the railroads, this idea that once you've connected it, then everything takes off, then it becomes transformational. And I think that there's a similar fantasy here, that once we hit the singularity or that point of exponential growth, then all of this will have been worth it. Which is bullshit, but I think it is part of what is continuing to drive the bubble here.

Matt Scherer: I do think that we're going to hit a singularity, but it's not the singularity that the tech bros think it is. What I've actually been talking about with people is if there's a deep enough downturn in the AI sector, the scale of the bet on AI has become so big that it is really hard to predict what will happen, if it bursts and the downturn is deep enough. There will be an economic singularity past which I don't think anybody who, no matter what economic and finance tools they have at their disposal, can really predict what happens when a $7 trillion CapEx cycle ends up essentially setting $7 trillion on fire. That's never happened before. We don't have a precedent for what's gonna happen in the economy in that instance. So that's the singularity I'm worried about. 

Alex Hanna: It's a pretty chilling singularity. I am not looking forward to that one either. 

Emily M. Bender: No. Should we get to this one here about "The Costs Are Here, The Returns Are Not"? This is the graph that we've been talking about. 

Alex Hanna: So this is, I think we've mentioned this. So, "The evidence is hard to ignore. The Financial Times recently reported that Amazon, Alphabet, Microsoft, and Meta are on course to spend-" sorry, what now? 

Emily M. Bender: Aamm! A-A-M-M. 

Alex Hanna: Oh yeah. Or I think you could also abbreviate this as ma'am. 

Emily M. Bender: Yeah, or mama. 

Alex Hanna: Or mama! "-spend 720 billion on AI infrastructure. The spending splurge would push their combined free cash flow, the cash left over after an expensive investment, to a decade low. The report projected combined free cash flow of $4 billion in the third quarter of 2026. That's compared with a post-pandemic quarterly average of 45 billion." So they are a little data center poor. "Put plainly, some of the world's most cash-rich software companies are starting to look a lot like heavy industry. No longer are they just writing code. They are buying chips, land, power, and data centers at enormous scale."

Emily M. Bender: And arguably they're actually not writing that much code anymore, because they're making people use these stupid systems to vibe code for them. 

Alex Hanna: Yeah, and I guess arguably they're also not necessarily directly building these infrastructures, but they're doing it through third party-

Matt Scherer: Lots of off-balance-sheet debt.

Alex Hanna: Whatever weird other stuff's happening there. Well, talk about the debt here too, because your report goes into the debt and the ways that- so the way that they're able to effectively build all these is because they're leveraging their valuations for all this private AI funding that they're getting. And this chart shows this, but your report goes into it a bit too.

Matt Scherer: Yeah, and it's interesting. A year ago, if we were having this conversation, there wasn't a lot of debt that had built up yet around the AI bubble. It was still at the point where the tens of billions of dollars in revenues that the big tech giants were bringing in were able to cover the CapEx that was being spent on this. But that broke around the summer of last year. And it's really, like it's stunning if you look at some of the graphs about how the debt just skyrocketed starting in the fall of last year. Oracle, I think, has taken on something like 100 plus billion dollars in debt just in the past nine to ten months. And same with Amazon, Google, Meta. Microsoft has, at least on its balance sheet, not taken on as much debt as the others. But the thing that is most worrying to me, just like reading a lot of the stories about this, is the number of echoes of Enron and the subprime bubble, where there are all these stories about ways in which debt is accumulating not on the company's balance sheets. Where they are setting up these shell companies and using different financial, mechanisms to make debt appear as not debt, or to hide it. And that is always- that's never been a good sign, ever. When you have to use those mechanisms as much as these companies are to hide how much you are taking on in debt, that is a sign that you are not confident that you're gonna be able to meet those debts. 

Alex Hanna: Yeah, 100%. And that's already after all the very creative restructuring work these companies are doing. Oracle just had a set of record layoffs and has attributed some of those to, quote-unquote, "AI." But more of it, where you're seeing who they're laying off- I forgot what the reporting was, but it might've been in Wired, but I'm not sure. But they were interviewing the people who they were laying off, and a lot of them were senior engineers who were about to have a bunch of money vest, and then were laying them off right before the vesting cliff. 

Matt Scherer: An old Henry Ford move. 

Alex Hanna: Yeah. Classic move. They have also been requiring AI usage. They've installed more bossware on machines to ensure and try to enforce that usage. So a lot of these types of things that are supposed to not only free up revenue for more AI investment, but then also signal to investors that they're making this move and are trying to increase valuations so that they can then leverage that into more CapEx.

Emily M. Bender: Yeah. So I usually do my very best to ignore the ads that pop up. But there's one on screen right now which is so on the nose. So the image looks like it's probably extruded. But it's these two guys, looking like they're from the time of the Gold Rush, and there's this little hand-painted sign on the side of a shed that says, "Picks and shovels." And then the one guy with a white beard is holding a pick, and the other guy is looking like he's maybe ready to receive it. This is part of why I think this is extruded, is it doesn't quite make sense. But the tagline is, "The best investors see the bigger picture." Which leads me to ask, like, where is NVIDIA in all of this? Because they're the ones selling the picks and shovels. 

Alex Hanna: And it was a Blackstone ad, if anyone's interested.

Emily M. Bender: All right, so more slop there. But NVIDIA, as I've understood it, like NVIDIA and then also the people selling access to other people's labor in terms of data annotation, are the only ones making real money in this. But I've also heard that NVIDIA is involved in a lot of this circular debt situation because they're lending money to these companies so that the companies can buy their chips. What do they look like right now? 

Matt Scherer: Right, well, I think everybody's revenues, including NVIDIA's, including the chip makers', are somewhat illusory. And what I mean by that is that NVIDIA, there is a great piece on this by Michael Burry, the investor who was featured in "The Big Short." But he pointed out NVIDIA has $180 billion in essentially, "We have contracted with TSMC to pay you this much regardless of whether you build our chips or not," because TSMC has orders so far in the future that they're not willing to reserve factory space, fab space essentially, unless they get hard contractual commitments. Now, that $180 billion spending commitment is not like a loan. It's not a traditional form of debt. But it is debt. And NVIDIA, as much money as it's making, $180 billion- that is like double its annual cash flows. So like, it is not the case that anybody is making profits that I would call secure, because ultimately it's all based on the big tech giants using, initially their cash flows, but now increasingly debt, to buy chips, which the chip makers are using, or the chip designer NVIDIA is using, to commit to buying more chips from TSMC after they fabricate them. And it's all based on this very fragile Jenga tower that at the bottom will ultimately depend on OpenAI, SpaceX, Anthropic, and Google, Microsoft, and anybody else who's selling AI models earning enough money actually selling AI goods and services to pay it off. So sorry, that wasn't actually your original question, Emily, about NVIDIA.

Emily M. Bender: No, but you got to it. And I just want to add to that whole chain of things is that these chips don't just function in isolation. They have to go into the data centers that nobody wants in their backyard- which is great. We're seeing lots of really good resistance to data centers. But that's another place where, okay, so NVIDIA is committed to paying for these chips that may or may not get made in the hopes of selling them to the companies that may or may not actually have anywhere they can put them.

Matt Scherer: And the data centers are not things that can be easily repurposed into other uses. They can't even be repurposed into traditional cloud storage and infrastructure data centers. They are designed to house NVIDIA chips and usually specific generations of NVIDIA chips. And I had a great conversation with Mitra Ebadolahi from Upturn, I remember, and some other folks last year about what are we gonna do with all these AI data centers once they get mothballed? Because, nobody- there's not enough demand for it anymore. Her idea was roller skating rinks, which I really liked.

Alex Hanna: Oh, as a roller derby player, I love it. We could actually see a complete, we could have really big tournament spaces. I would probably want them to actually have some housing or something, though. 

Emily M. Bender: Yeah. There's probably enough room for both, 'cause you're talking about the one that's the size of Manhattan.

Alex Hanna: Oh yeah, completely. 

Matt Scherer: The one in the Utah desert, yeah. 

Alex Hanna: Yeah, exactly. You just have to move to the Utah desert. Wow, okay. So this is getting to the revenue piece. So this is the piece by David Cahn, "AI's $600 Billion Question." And then they start talking about the revenues. And just to read this, what these would anticipate, so, "OpenAI illustrates the strain pinching the industry. The Financial Times reported this spring that OpenAI secured up to 110 billion in funding at a $730 billion valuation. The same reporting tied that funding to huge compute commitments, including a $100 billion Amazon agreement and about 600 billion in compute commitments through 2030. A valuation that large requires an extraordinary outcome. OpenAI would need to be one of the most profitable companies in history while also paying for some of the most expensive infrastructure in history. That could happen." Which is... what a thing to say. "But it is an enormous undertaking. "Anthropic offers the strongest reason for optimism. The Financial Times reported that the company expects a profitable quarter in 2026-" so it hasn't had one yet- "with revenue above $10 billion in the second quarter. That is real evidence of demand. The same reporting suggested Anthropic's valuation could reach 900 billion, while the company recently signed a $15 billion annual compute commitment to SpaceX. A profitable quarter is an important milestone, but it does not prove that today's prices for the whole AI sector make sense." Okay, thoughts on that?

Matt Scherer: Going back up to that graph, like- okay, so Anthropic's gonna make $10 billion this quarter. Let's say that it sustains that and it makes $40 billion over the course of the full year. That does not- 

Emily M. Bender: Hold on. That's still like here... 

Matt Scherer: Exactly. Yes. 

Alex Hanna: That's still very short.

Matt Scherer: Very short. More than an order of magnitude below what is needed to cover, again, that left graph, just six companies' spending commitments on AI. 

Emily M. Bender: And that is the plan for 2026, right? So this middle one, this is "private AI funding since early 2024," is about half that amount, and then the application spend, presumably across all companies, is at 19 billion. So your imagined 40 billion, let's say more, still only gets us to 59 billion, and this thing over here was 725 billion. So yeah, that's not... 

Matt Scherer: And the assumption is like, maybe they're a growing company. Maybe those revenues will grow even faster. But that is, again, it's like a very optimistic memory. It's like, hey, Amazon kept making more and more money as time went on, so maybe Anthropic will make more and more money as time went on. But Amazon's business model, number one, it was built on exploiting labor, exploiting workers, exploiting third-party sellers and stealing their ideas and a million other things. But number two, it also was based on, quite frankly, a much more proven idea. Amazon was not selling things that there was no evidence of whether they worked or not or whether people wanted it or not. They were selling things that we knew people wanted, which is why Amazon was selling them. They were simply trying to sell goods that already were being sold by other companies, just at a lower price. That's not what Anthropic is doing. It is essentially trying to build a new market, on a technology whose, the number of profitable commercial applications for it is highly speculative at this point. And I would argue that, I agree with you that the inherent limitations of large language models and of generative AI systems make it very difficult to envision a future where it's going to reach anywhere close to the profits that these companies are imagining. 

Alex Hanna: Yeah, and I did wanna say something about Amazon. Amazon also made quite a bit of money on their, just on building out a huge logistics system and using it to undercut a lot of brick-and-mortar retailers. So they were willing to be a bit of a loss leader in terms of their shipping, but they were able to make it up in volume, right? And then- 

Matt Scherer: This is awesome. I haven't seen this webpage before.

Emily M. Bender: So mjkranz in the comments suggested it, and it's reminiscent of "Has the large hadron collider destroyed the world yet dot com," which just displayed no. But this is more dynamic. So it says, "Is AI profitable yet?" The URL is isaiprofitable.com. And subhead here, "Tracking the spend and revenue of frontier AI companies, May 2026." And the answer is "No, everyone's broke." And the most alarming thing on this is there's this counter going for dollars spent on AI since page load. So since the time I've loaded it, like by the time I'm done saying this, we're gonna hit 2.4 million. That's just an alarming pace of spend. It's a little surprising to me actually that OpenAI is not further in the red here. But that might have to do with some of that hidden debt that you're talking about, Matt.

Matt Scherer: It's not hidden debt. Well, they're getting other people to spend money essentially on their behalf. 

Alex Hanna: Yeah. That's what it is. It's effectively, you're just having these deals that you have with other people who are building your CapEx, right? Or they're building your capital assets. 

Matt Scherer: If you added OpenAI's lease commitments, which again are essentially a form of debt, but it's not one that shows up on your current balance sheet necessarily- You know, that's a question for another day, whether lease commitments should be treated as a form of debt in these situations. But if you count OpenAI's lease commitments, that red bar would be out near, at the very least, where Meta's is. 

Emily M. Bender: Got it, yeah.

Alex Hanna: One of the things I wanted to raise up, there was some very funny stuff in the chat about just the- magidin says, "Is it evaluation or is it bragging wishcasting?" I replied, "It's only bragging wishcasting if it comes from Santa Clara County." And magidin says, "Sparkling lies, then." And then, and also says, "Otherwise their favorite grapheme to indicate, quote, 'AI'? Yep, sparkling lies." Given the sparkle emoji.

Emily M. Bender: Perfect, perfect. Is there anything else we wanna do in this artifact, or should we move over to our second one? 

Alex Hanna: I'm kinda looking at the bottom. I'm actually interested how this person ends, which isn't optimistic at all, 'cause he- 

Matt Scherer: I actually found it fascinating, sorry, the section that you have up on the screen right now. 'Cause again, this is a guy who was at the Mercatus Center, which is at George Mason and is essentially a libertarian institute. And he was at the Competitive Enterprise Institute, which has close ties with the tech industry. And here he is acknowledging a truth that I point to in my report. And it just points out that the whole idea that Silicon Valley is responsible for all of this technological innovation that we're seeing, it's bullshit. It's taxpayer-funded programs, and not just taxpayer-funded programs in the US, but also the work of academic researchers in the Soviet Union and Japan, and lots of other people who worked at CERN. That's where Tim Berners-Lee invented the World Wide Web, essentially. I just found it very interesting, and kudos to him for acknowledging that this, the whole idea that the digital revolution that we've experienced and the technologies that formed the foundation for AI today- that did not come from Silicon Valley for the most part. That came from people who were doing research in the public interest and often using taxpayer funds.

Alex Hanna: 100%. Yeah, I think this is more of the same at the end. Maybe we should go to the next item.

Emily M. Bender: Yes. Okay, so this is from CNBC. Journalist is MacKenzie Sigalos from, it's a little bit older, August 18th, 2025. And the headline is, "Analysts downplay AI bubble worries as Altman says some investors will be left, quote, 'very burnt.'" And we're gonna skip the key points 'cause those are probably synthetic. And it starts with, "The artificial intelligence boom that Sam Altman helped ignite with ChatGPT in late 2022 is starting to make even him uneasy. Startups with little more than a pitch deck are raising hundreds of millions. Valuations have become, quote, 'insane.' Capital is chasing, quote, 'a kernel of truth' with feverish speed. The OpenAI CEO still believes the long-term societal upside of AI will outweigh the froth, and he's ready to keep spending in pursuit of that goal. Quote, 'Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes,' he said at a recent dinner with reporters. 'Is AI the most important thing to happen in a very long time? My opinion is also yes.'" And then he says, the next bit is, "He repeated the word bubble three times in 15 seconds, then half-joked, 'I'm sure someone's gonna write some sensational headline about that. I wish you wouldn't, but that's fine.'" 

Alex Hanna: I mean Sam Altman always comes across as the least media-trained individual in this industry. But I guess if you are the head of this damned company, you get to say any kind of bullshit.

Matt Scherer: It's like the ultimate kind of "Aw, shucks," you know, is how he comes off. Despite the fact that anybody who's read "Empire of AI" or any of the other exposés on him knows that he's anything but an innocent, "aw, shucks" dude.

Emily M. Bender: Yeah, absolutely. So this next bit is, I think, also worth reading just because he continues to be punchable. "'You should expect OpenAI to spend trillions of dollars on data center construction in the not very distant future,' Altman said. 'And you should expect a bunch of economists wringing their hands saying, "This is so crazy. It's so reckless." And we'll just be like, "You know what? Let's do our thing."'" 

Matt Scherer: Okay. So sorry, this is the part that when I got to earlier, like, blood pressure spiked the most. This is the difference between this bubble and prior bubbles. Before, sure, it's like, okay, if you spend a few billion dollars on a tech that doesn't live up to its hype, some investors will lose their shirts, but most people aren't gonna notice. Think of Facebook changing its name to Meta and going all in on AR/VR. Yeah, that sucked for Meta. Most people, that didn't affect their lives in any meaningful way. But when you're talking trillions of dollars, that's no longer an amount of money that you can afford to more or less flush down the toilet and say, "Let us do our thing." Because at that point, everybody's gonna feel it if the bet doesn't pay off. You're not just gambling your shareholders' money. You're basically putting the stability of the economy on the table. 

Alex Hanna: Yeah, 100%. sjaylett says in the chat, "'Let us do our thing,' says man setting fire to other people's houses." That sums it up. 

Emily M. Bender: And above that, we've got abstract_tesseract: "Are we in a bubble? Glances down at magic 8 ball. Signs point to yes." And Blend3rman says, "Sam's 8 ball must be all affirmatives." ChatGPT is, and that's Sam's 8 ball, right? All right. Should we briefly check in with Bezos here?

Alex Hanna: Let's go to Bezos, yeah, 'cause the rest of this isn't saying much. So this is more recent, also from CNBC, not the same journalist. By Annie Palmer, May 20th, 2026. And the headline is, "Bezos brushes off concerns of an AI bubble, quote, 'You shouldn't worry about it.'" And this is Bezos, who is sporting a little bit of scruff these days. It's very interesting that he's gone the scruff route. And he told CNBC, quote, "Even if it does turn out to be a bubble, you shouldn't worry about it, because the bubble is driving investment and a lot of the investment is going to turn out to be very healthy." Is that how bubbles work? 

Matt Scherer: It is not. And it's interesting, he told that to Andrew Ross Sorkin, and I would highly recommend two of Andrew Ross Sorkin's books to help explain why this is bullshit. "1929," which came out this past year and provides a really interesting in-depth look at how the stock market crash of 1929 occurred and snowballed into the Great Depression. As well as "Too Big to Fail," which did something similar with the financial crisis in '08. But the 1929 stock market bubble, like the railroads, that was an example of a bubble that was driven by technologies that had real society-changing potential and reality, like radio, aviation, consumer electronics. And eventually, companies did figure out how to profit from those technologies, and eventually they did transform society, in the long run. But to quote John Maynard Keynes, "In the long run, we're all dead." The fact that a technology eventually, after the bubble bursts, will produce profits and societal change, that wasn't much comfort to people who were living in Hoovervilles.

Emily M. Bender: Right. So, given that we can see a bubble burst coming, and given that we know that the first impulse of our government is to try to shore up the companies instead of do something more equitable, what do we do now?

Matt Scherer: I'm honestly struggling a bit with that. I think number one, we do need to make a lot of noise preemptively about, there shouldn't be bailouts. And however much sense a bailout might- you can argue, and I would, about the merits of the bailouts in the wake of the 2008 financial crisis. Those, I think, were a mistake. But they were more justifiable, I would argue, in a sense, than bailing out tech companies that are spending this much money on AI. Because number one, I just don't see it... Again, GPUs are not houses. The houses that were built at the peak of the housing bubble, people can still live in them. We're not gonna be using these GPUs in a few years. So essentially, you're just giving money to people who had already set that money on fire by bailing out the tech companies this time around. Beyond that, I would say that we should start preparing for ways to take the assets that are left behind in a bubble and convert them into more worker, consumer, and taxpayer-friendly purposes. There's no reason, and there's a report from OMI that went into this a year or two ago, about how cloud infrastructure, which is essentially the infrastructure that runs the modern internet and the digital economy. That should be a utility. And the bursting of the bubble provides an opportunity to take some of that infrastructure and make it so that people have easier access at non-monopolistic prices and terms to the basic infrastructure that you need to run a business and to just exist in the modern world. So I think that we need to start thinking about those things. I just don't have enough confidence in the Trump administration or in Congress that they will do anything to actually prevent the bubble from bursting before it bursts. We should start thinking about how it is that we can take the useful pieces that are left behind in the bubble, which unfortunately are a lot less than the useful pieces that were left behind in past bubbles, but find a way to turn them into something that actually benefits consumers and workers.

Alex Hanna: Yeah, absolutely. And I think one of the things, I did wanna read this just because I think one of the myths of this between both Altman and Bezos is that the technology is useful. And not only that it's going to take off, but he says in this article, "Bezos acknowledged the excitement around AI has meant that, quote, 'Every experiment is getting funded,' end quote, including potentially bad ideas. Quote, 'It's because investors at this moment haven't learned yet how to discriminate between good ideas and bad ideas, and that's okay because the good ideas will pay for all of the losers,' Bezos said. Quote, 'So from the point of view of civilization, of society, these kinds of industrial cycles can actually be very healthy, because they drive the technology forward,' end quote."

Emily M. Bender: Get in, losers, we're burning cash. 

Matt Scherer: Yeah. He keeps using the words "very healthy." It's kind of like Trump's medical reports. You're protesting a little bit too much about how healthy this all is. 

Alex Hanna: Yeah, exactly. There's a lot here to dissect, both in terms of what is considered healthy, what is "point of view of society," end quote, which is very loaded in terms of, what society? What civilization? And who are the losers here, that are non-investors? Who are the people who are gonna lose their shirts? There's a bit of chat in the chat about the kinds of ways after 2008 in which the people that really needed the money were the people who lost their homes, the people who are the most marginalized, and really could have used that to rebuild, rather than bailing out the banks and whoever else.

Emily M. Bender: Yeah. So Alex, I'm taking us back here because I'm using it for inspiration for the transition. 

Alex Hanna: Ooh, okay. 

Emily M. Bender: Do you want musical or not musical? 

Alex Hanna: We should try musical. I feel inspired today. 

Emily M. Bender: All right. I need a style that allows for frenetic energy. 

Alex Hanna: Oh, I'm gonna scat. 

Emily M. Bender: Okay. You are in charge of updating these graphs and painting the red bars as they grow. And you're singing about that. 

Alex Hanna: Okay. I've been listening to "The Scatman." You remember that song? But so, I'm the red paint guy. Painting red lines on the way. I'm gonna give it, gonna do what to do. If I can paint red lines, and the Scatman can, so can you. I'm out of green paint, I'm ready to rhyme, and I'm ready to get it going. Except we're going down the drain of an AI bubble, but I am gonna keep on flowing. That's all I got.

Matt Scherer: Impressive. 

Emily M. Bender: You knocked that out of the park. 

Alex Hanna: Thank you, that was- I felt a lot of pressure. 

Emily M. Bender: That brings us to our Fresh AI Hell, if I can get organized and share it. So the first item is some reporting, so it's on Bluesky from Phil Lewis, but it's a link to Action News 2 from Atlanta, and this is from May 14th. And Phil Lewis writes: "Dozens of empty Waymos invaded an Atlanta neighborhood and circled a cul-de-sac for hours with no passengers." The way of the future for sure. 

Alex Hanna: Very good. So this is from Ana Marie Cox, a journalist, it is also a skeet. And she says, "This is legitimately one of the most horrifying things I've ever seen in all the LLM bullshit that's out there. That, quote, 'joyful' is misspelled seems especially meaningful." And it is a screenshot of a tool called AA Digital Sponsor. Oof. So, "An old, friendly, and well-read sponsor for people in the recovery program of Alcoholics Anonymous. This module is designed to provide a listening ear and coach people in thinking through the spiritual guidance of the Big Book." This is horrifying. Especially if you know somebody who's going through recovery, the idea of a sponsor is to really aid you in navigating your recovery, and also telling you some hard truths and also going to social places and being in a room with people. So this is, yeah, truly fucked up. I agree with Ana Marie here.

Matt Scherer: I saw not one, but two, but three Ls in joyful, on that. 

Alex Hanna: Oh, I didn't even see that. Wow. 

Emily M. Bender: Oof. This is a tweet by Dr. Damien P. Williams- not tweet, a skeet- from May 18th, 2026, quoting a skeet from The Verge. And the headline there is, "Chatbots at the drive-thru are just the beginning." And The Verge writes, "Taco Bell is experimenting with an AI-driven menu board, which will use the technology to, quote, 'dynamically change the layout, content, and visuals on a car-by-car basis.'" And Dr. Damien P. Williams says, "Having some bullshit generator randomly changing the layout of the menu so you can't find shit is literally the opposite of what a fast food experience is supposed to be." And then he adds, "And you're gonna do this at Taco Bell, where at least one passenger in any given car has a greater than 40% likelihood of being high?"

Alex Hanna: Yeah. Wow. 

Matt Scherer: And I live in Oregon, where that would be an underestimate.

Emily M. Bender: So we have a friend of the show, Arturo Magidin. You wanna do it, Alex?

Alex Hanna: Yeah, hi! You're in the chat, and now you're on the screen. So Arturo says, "YouGov is polling on whether people want to have a chatbot be a juror. What fresh AI hell is this?" And there is the screenshot of the poll is, from a YouGov survey, and it says, "If you were falsely accused of a crime and it went to trial, which of the following would you prefer to listen to the arguments and give the verdict?" The options are "a jury of your peers," "the best widely available AI chatbot," and then, "not sure." I'm curious on what the results of that will be. 

Matt Scherer: Yeah. I'm curious as to who commissioned that question. 

Emily M. Bender: Oof. Yeah. This is actually, should've saved it for a little bit 'cause it is a chaser. Erin Brockovich, that Erin Brockovich, is on the case. She has started something called Brockovich AI Data Center Reporting, which is fantastic. And there's a quote from her: "The race to build AI infrastructures is unfolding town by town across America. In some places, data centers are welcomed. In others, they are delayed, contested, or abandoned altogether. This map captures the real world footprint of that race, revealing patterns of growth, conflict, and uncertainty. I am watching as you, the communities, show up and speak out. In the famous words of Mark Twain, 'The secret of getting ahead is getting started.' So let's go." So, that's exciting. 

Matt Scherer: I have to say, my real claim to fame is I went to the same elementary school as Julia Roberts and the same high school as Usher. 

Alex Hanna: Wow. We just need Usher to get on the case.

Matt Scherer: I'm afraid I don't have an in with him.

Emily M. Bender: Usher, if you're listening. Okay, Alex, you can have this one while I try to dig up that other link. 

Alex Hanna: Okay. Oh gosh, you're gonna make me say this on the pod? 

Emily M. Bender: Do you want me to do it? I can do it. 

Alex Hanna: No, I'll do it. So this is decrypt.co, and the title is "AI Startup Says It Will Pay People $2,000 a Month to Masturbate. Yes, really. Joi-" spelled J-O-I, oof, I just got that- "AI is hiring 10, quote, 'masturbation consultants' to test its AI-guided masturbation feature and report how it affects stress, sleep, mood, and confidence." This is by Jason Nelson, May 24th, 2026. And then scroll down, there's a picture of a stack of money. Let's see if there's anything else to this that is worth saying. "Quote, 'The role is real, and we've had great responses since the posting went live,' Joi AI Head of Brand and Communication Julie Levin told the Decrypt." And then, it's a screenshot or an embedded, from Twitter, the ad of getting hired there. 

Emily M. Bender: Yeah. So this is ridiculous. Basically, they want people to use this app and share a bunch of data. So, "The listing describes ideal candidates as articulate, observant, and impossible to blush," which is ungrammatical. But, and then, "people who can describe sensations, quote, 'better than a sommelier describes a wine.'" 

Alex Hanna: Oh my gosh. Do you know what J-O-I stands for? I didn't know this until- but it stands for jack-off instruction. 

Emily M. Bender: Oh, no. 

Alex Hanna: And so that's why the name is both joy, but it's like- it's terrible. I'm just, I'm appalled. 

Emily M. Bender: Yeah. Ugh. And it continues here, "The posting also promises flexible scheduling and, quote, 'the most interesting what do you do for a living answer at any party.'" All right, I couldn't get to the archived version of this Wired article, but I do have Carl Bergstrom's post about it on Bluesky. So his remark on May 17th of 2026 is, "Oh, good grief. No, they don't." And the headline from Wired is, "Overworked AI Agents Turn Marxist, Researchers Find." Subhead: "In a recent experiment, mistreated AI agents started grumbling about inequality and calling for collective bargaining rights." And when I clicked through and read the thing, it was of course Stanford. It was a political scientist and a couple of economists from Stanford, who played with some interactive fiction and got excited. 

Alex Hanna: Oh, who was it? I need to look this up later. 

Emily M. Bender: We might be able to get that far, at least, before we hit the paywall. Eh, not sure. You'll have to look it up later. All right. And Alex, you wanted this one, so you get to read it. 

Alex Hanna: Yes, okay. So Anthony Moser, friend of the pod. Fantastic skeet thread, just one of the best, I think. He's a great storyteller. So I'm gonna read this in full. "Tonight in the AI reading group, I was reminded of something that happened in 2006 after I dropped out of college, when I was living in a rundown old house next to J.B. Alberto's in Rogers Park with a bunch of dudes. It was, as you can imagine, not clean, and I had this roommate who did not help with the chores, certainly. More than one, actually. But this guy, I came into the kitchen one afternoon and I was shocked. It looked great. He had done all the dishes, all the glasses, sink was clean. I honestly didn't think he had it in him. So I get a cup. I'm filling it up in the sink. I'm like, 'Man, thank you so much. This is amazing.' And then I notice the scrub brush on the edge of the sink. Which looked extremely, distressingly similar to the scrub brush we kept under the sink with the toilet bowl cleaner to clean the toilet. With a feeling of both certainty and dread welling in my stomach, I asked him where he'd gotten the cleaning supplies, but I already knew. I poured out the cup, threw away the brush, and from then on trusted nothing I had not washed myself immediately prior to use. His was a human error, but it occurred to me that it's the kind of mistake you get from AI. Because the form was perfect. He seemingly did the thing, used the tools, went through the motions, produced exactly what you would expect to see. But there was a key piece of context missing, and it completely changed the meaning of his actions. You have to understand why you're doing the things. You have to understand where the scrub brush comes from. And that feeling of, 'Oh no,' then visceral revulsion settling into deep distrust, that's how it feels when someone gives me slop. You just handed me a cup of water, and I'm looking at the sink going, 'Where did you find that brush?' In a way, it's actually worse. At least he didn't realize he was doing the dishes with the toilet brush. People handing out slop know that they're using the toilet brush. They're fine with it. 'Hey, this brush works great on everything. Here, have a cup of water.'" 

Emily M. Bender: Ew! 

Alex Hanna: I am like, oh, the storytelling. This was like a mini horror novel with a great message, a horror thread. Just, so good. I love this as a storytelling exercise. 

Emily M. Bender: Yeah, really great. And we put it in as a chaser because it is a good kind of pushback against all of this nonsense. All right, we made it. 

Alex Hanna: Well, that's it for this week. Matt Scherer is a fellow at the anti-monopoly think tank, Open Market Institute. His latest report is "No Bailout for Tech: Big Tech Billionaires' Policies for When the AI Bubble Bursts." Thanks for coming, Matt. 

Matt Scherer: Thanks so much for having me.

Emily M. Bender: Thank you. It's been great having this conversation with you. 

Alex Hanna: Our theme song is by Toby Menon. Graphic design by Naomi Pleasure-Park. Production by Ozzy Llinas Goodman. And thanks as always to the Distributed AI Research Institute. If you like this show, you can support us in so many ways. Order "The AI Con" at thecon.ai or wherever you get your books, or request it at your local library. 

Emily M. Bender: But wait, there's more. Rate and review us on your podcast app, subscribe to the Mystery AI Hype Theater 3000 newsletter on Buttondown for more anti-hype analysis, or donate to DAIR at dair-institute.org. You can find our merch store there, too. That's dair-institute.org. You can find video version of our podcast episodes on Peertube, and you can watch and comment on the show while it's happening live on our Twitch stream. That's twitch.tv/dair_institute. Again, that's dair_institute. I'm Emily M. Bender. 

Alex Hanna: And I'm Alex Hanna. Stay out of AI Hell, y'all.

Head Shot

Alex Hanna

Co-host
Head Shot

Emily M. Bender

Co-host