May 28, 2026
On May 28, The Capitol Forum held a conference call with Matthew Scherer of the Open Markets Institute to discuss his recent report, “No Bailouts for Big Tech Billionaires: Policies for When the AI Bubble Bursts.” The full transcript, which has been modified slightly for accuracy, can be found below.
TEDDY DOWNEY: Hello, everyone and welcome. I’m Teddy Downey, Executive Editor here at The Capitol Forum. Today, I’m very pleased to be joined by Matthew Scherer, a fellow at the Open Markets Institute, whose work focuses on technology policy, market concentration, and the political economy of artificial intelligence.
Matthew is the author of the recent report, “No Bailouts for Big Tech Billionaires: Policies for When the AI Bubble Bursts.” It’s an awesome paper. I really enjoyed reading it and I’m super excited to talk to Matthew. Matthew, thanks so much for doing this today.
MATTHEW SCHERER: Thanks very much for having me.
TEDDY DOWNEY: So, the paper is done in several stages. I’d like to stick with that logic. And first, you argue we’re in an AI bubble. I think there’s some disagreement about that. I would love to get your thoughts on how you came to that conclusion. What do you think is the strongest evidence that we’re currently in an AI speculative bubble?
MATTHEW SCHERER: For sure. I think that there’s two things to bear in mind about that. Number one, I think a lot of folks think that if AI is a transformative technology and a revolutionary technology, that there’s somehow a disconnect or a contradiction in saying that it’s possible for there also to be a speculative bubble around it. And that’s just not true.
I always start with kind of this just little historical reminder that a lot of people aren’t familiar with, which is that the largest and most damaging speculative bubble in history was the railroad bubble in the 1870s. And when it burst, it led to the longest depression in the history of the United States. And it wasn’t because railroads weren’t a revolutionary technology. Railroads were a revolutionary technology. It probably was the most significant revolution in logistics and transportation that the country has ever experienced. It was the first time that motorized transport between cities over long distances became available. But the issue was that there were too many people building too many railroad tracks too quickly. And investors kind of got a little exuberant and they built way more than there was capacity to use at the time.
So, that is kind of one historical precedent that I’m bearing in mind for this. And there’s a lot of the same dynamics at play today. I’m not going to lie, I’m somewhat of a skeptic on AI’s revolutionary potential. I think that there’s a lot of hype around AI’s capabilities and its ability to transform the economy that is speculative at best.
Right now, we are not seeing any productivity gains from AI in the macroeconomic statistics. Right now, there are plenty of studies out there, independent studies, showing that there are—not no, but very few—individual companies that are actually seeing productivity or profitability gains due to AI, other than for the companies that are selling cloud services for AI.
But regardless of whether you think that might change and that down the road there’s going to be some sort of profit, the problem is that right now, the stock market is more concentrated in a handful of companies, in a single sector, than it has ever been in history. Right now, the top eight, I believe—yeah, I believe it’s still eight—companies in the S&P 500 are all tech companies that are deeply invested in the AI bubble, and they make up a good third to 40 percent of the entire S&P 500. That’s unprecedented, that sort of concentration, in essentially a bet on a single technology. The stock market, at the same time, is larger relative to the economy as a whole than it has ever been. So, there is a lot of money just riding on the stock market, is the first aspect.
TEDDY DOWNEY: Can you stay on that for a second? The percentage of GDP in the stock market, that was an interesting stat. I hadn’t seen that before. How did you come up with that? And why do you think that is indicative of us being in a bubble? How does that compare historically?
MATTHEW SCHERER: So, it’s interesting. The ratio of the total value of the stock market to GDP is actually called the Buffett Indicator. At least colloquially, that’s what it’s called. You can imagine why it’s called that. It’s because Warren Buffett once said that it is something that is important to keep an eye on. He was talking a couple of years after the bursting of the dot-com bubble. And he pointed out that at that time, the stock market, the ratio of stock to GDP was about, I believe at the time, 150 percent, which was higher than it had ever been. And he said that that should have been a big warning sign that there was frothiness and that there was kind of overinvestment. So, that was in 2001, it was 150 percent.
Today, that metric, the last time I looked, is around 220 percent. So, half again bigger than it was when Warren Buffett first made the warning that gave rise to the name of the Buffett Indicator. And that doesn’t even count the fact that we now have the largest privately held companies in history, SpaceX, Anthropic, and OpenAI, which are all admittedly about to go public, which will drive the Buffett Indicator even higher.
But the point is, right now, the total amount of money that is tied up in what are called equities—basically stocks, shares in companies—is about three and a half times GDP. And that is just leaps and bounds more than it has ever been on record.
So, most of that is held by very wealthy people because of wealth inequality, and wealthy people tend to own more shares of investments. But that shouldn’t be much comfort because our economy has become increasingly dependent on wealthy people spending money.
So, we’ve got the stock market being bigger than ever and more dependent on a single sector than ever. And that’s not even getting into the massive amounts of debt that has been building up, especially over the past about nine months or so.
TEDDY DOWNEY: And can you talk a little bit more about just the indicia of being in a bubble? You mentioned in the paper, massive amounts of debt, hiding the effort to hide the debt, the circular accounting shenanigans, all the steps that are going on that are atypical, really, in a healthy market.
MATTHEW SCHERER: So, there’s actually a book by Brent Goldfarb and David Kirsch that I cite in the report called “Bubbles and Crashes”. And they say that there are four indicia to tell you that you’re in a bubble. Now, they don’t present it as some sort of iron law of economics, but these are things that if they are in place, they strongly suggest that the dynamics of a bubble have kind of taken over economic reasoning.
And the four factors are you have the ability to invest in pure-play companies, in other words, companies that are kind of specifically devoted to whatever the technology is at the center of the bubble. You definitely have that today with OpenAI, Anthropic, and lots of startups like—or not startups, but companies that people are less familiar with like CoreWeave and Nebius that are specifically devoted to providing infrastructure to AI companies.
You have uncertainty—that’s the second factor—about how the technology will play out, what its economic benefits will be, and how quickly those benefits will come about.
You have the presence of lots of novice investors, which the examples that I kind of point to as the obvious ones are Kevin O’Leary of Shark Tank, starting essentially an AI cloud infrastructure startup. Donald Trump’s son’s doing the same. And Allbirds, the shoe company that makes eco-friendly footwear, making a pivot to AI compute and calling itself NewBird AI, which you would think that investors would have just pointed and laughed at that, but instead their stock went up 600 percent. And that’s the sort of thing that really is kind of, I think five years from now, we’ll look back on that and be like, wow, we should have realized at that moment that things were getting out of hand.
Anyway, the last thing though, and the thing that I discuss at greatest length in the report is bubbles are driven by strong narratives. And the narrative around AI is the most compelling and seductive narrative that I think that there’s ever been around a bubble. They’re saying that AI can completely automate huge amounts of economic activity across the economy. And not just in the long run, but in the next few years, that we are imminently on the edge of an AI transformation.
And that has driven—really, if you look at a lot of the charts that are in my report and any other charts that kind of track the values of different sorts of tech companies, different types of debt that have been building up in the system. You see this just massive spike in things, either happening starting around the time that ChatGPT was released or with a lag of a few years after ChatGPT was released. And the risk is that the longer this bubble continues to inflate, the more money is at stake when it bursts and the more economic damage it can cause.
TEDDY DOWNEY: You also mentioned in the paper that one issue is that you need to have some idea of how these companies will make money, what the return will be, so that you can keep the whole thing going. Or otherwise, that would be an antidote to this in some way. Or it would be a refutation that, hey, there’s a lot of revenue coming down the pike other than this circular cloud infrastructure investment, cloud services.
And you make a point that I haven’t heard much, which is that the intention of a lot of these businesses is either to just funnel money into chatbots, how are you going to make money from that? It’s unclear. There’s not a lot of subscription revenue. We have no idea how they’re going to make money. Is it ads, really? When is that going to start happening?
And then the other is software to replace human workers, as opposed to tools to enhance productivity. I hadn’t heard—that’s an interesting argument that the business plans actually don’t seem to really make a lot of sense. Is that just kind of a tangential point? Is that just something that you noticed? Or how core is that to us being in a bubble?
MATTHEW SCHERER: I think that there’s two answers to that. The first is whether those revenue streams will be enough ever. And the second is whether those revenue streams, if they do materialize eventually, will materialize quickly enough to pay down the short-term debts that are being used to finance the construction of all these data centers to house specific generations of AI chips.
So, on the longer run question—as I said, I’m kind of more of a skeptic when it comes to AI’s economic potential. If you look at SpaceX’s prospectus recently, they kind of lean into the suggestion that the bet that AI model developers are making is that their models will be able to automate huge amounts of white collar and administrative work.
Now, I could spend a whole hour going into all the reasons why I very much doubt that that is ever going to be a viable prospect with large language models. They are simply too prone to errors.
One thing that I point folks to is that kind of quietly, Khanmigo, Khan Academy’s AI tutor program, was decommissioned last month. And the reality was that students and teachers simply did not find it useful enough, despite massive amounts of investment and partnerships with many different school districts and educational institutions. They just didn’t find it helpful enough to continue using it.
Uber’s COO said, you know what? We’ve spent tons of money burning through tokens on our AI infrastructure and we didn’t see a return on it. I think that that dynamic is playing out over and over again.
And there’s just this assumption that eventually they’ll figure it out, that AI will just keep getting better. But in reality—and this is a point that I kind of nod at in the report—the basic architecture of the large language models that are at the center of this bubble is not super-new technology. It’s not something that just came along when ChatGPT did. In a lot of ways, it’s an architecture that’s been around for 60 years and it’s been refined and they’re throwing more computational power than was ever possible at it before.
But there are a lot of folks out there in the computer science world who’ve made the argument that you’re going to kind of need fundamentally new breakthroughs in order to achieve that kind of white collar automation, generalized intelligence, that would be necessary to unlock—to use the number from SpaceX’s prospectus—$26 trillion, which is what they think the market size is for the automation of all these tasks.
But the problem is that the $7 trillion that is being spent on building AI infrastructure is not being built on finding these new foundational breakthroughs. It’s being built training bigger and better chatbots, basically. And I just don’t see the path to that ever earning back its investment, to be blunt, but certainly not earning back its investment by 2029, 2030, when a lot of these data center financing loans are going to become due in large numbers.
TEDDY DOWNEY: Before we move on to the next section of the paper, I wanted to bring up something that we’re seeing in the news, which is companies that have not a very important role in AI really, but that they’re now getting trillion dollar valuations, which are these legacy memory chip companies.
MATTHEW SCHERER: Micron.
TEDDY DOWNEY: Like Micron, where all they did, to my mind, starting in 2021, 2022, colluding to restrict supply. So, that if there was ever a demand spike, they would have tremendous pricing power. And then they’re now even saying, hey, we’re actually not going to build more capacity. We like this environment where you’re just begging us for chips and paying through the year. So, we’re going to keep you here.
And there are trillion dollar valuations. I mean, in some respects, I guess, if you see, well, there’s going to be no supply and tons of demand in perpetuity, okay. But how do you interpret the Microns and Sandisks and whatevers of the world that actually don’t really fundamentally have anything to do with the actual technology boom getting trillion dollar valuations?
MATTHEW SCHERER: Well, a couple of things. First, if I can give one piece of practical advice to anybody who’s watching this, if you’re going to need a new smartphone or laptop in the next couple of years, buy it now. Because the prices are going to go up a lot once the current inventory of devices that were built before the AI bubble started to suck in all of the memory chips kind of reached critical mass.
TEDDY DOWNEY: Yeah, they’re already going out. I mean, Nintendo, Sony, I saw that—what is the computer handheld one called?
MATTHEW SCHERER: Is that the Steam Deck?
TEDDY DOWNEY: The Steam Deck went up by $300, $200 yesterday. I saw an article about that. I mean, this is going to be such a big pocketbook issue already.
MATTHEW SCHERER: And people don’t realize how many different things in your life actually have memory chips in them within the era of, like, smart devices. Like, your car’s entertainment system has memory chips in it. You even have small memory chips in some wearable devices and things like that.
But it’s kind of interesting. Memory chips—it is not a new technology. In that sense, you’re right. It’s kind of incidental to the AI boom. But the fact that we are able to build memory devices that are capable of keeping track of as much information as they can in short-term memory is actually a major reason why you have such an explosion in capabilities of generative AI over the past few years. It’s simply because we’re able to throw all of this memory and compute at it.
That gives companies like Micron, just like it gives companies like Nvidia, kind of a stranglehold on key parts of the market. And you’ve got a Dutch optical fiber manufacturer—that nobody’s heard of—also that has a stranglehold on another part of this market.
And that’s another kind of aspect of the fragility of this bubble that I think a lot of folks aren’t quite keeping track of, which is that there’s so many different choke points. Where if a fab plant goes offline somewhere for a few months, there are so many bets that are being placed on each generation of large language models that are dependent on the delivery of the chips, the delivery of the memory, the construction of the data center, getting it connected to the grid, overcoming community opposition, and getting permits for everything.
And if any one of those things goes wrong, there is a lot of money riding each time on the successful completion of these projects within a relatively short timeframe. If there’s too long a delay, if the memory chips for whatever reason gets stuck in the pipeline, then the Nvidia’s chips that were designed to work with that memory will be sitting idle and new generations of chips will come out and supersede them before they ever come online.
So, there’s just all these different ways that this cycle could go wrong, in part because each generation of chips has a relatively short window of two to three years, maybe up to five or six, that they are going to be productive… But you have several years of lead time where the chips are either being warehoused or being fabricated or being designed, all of which has different contractual and lease commitments from multiple companies riding on everything happening relatively quickly and on time.
That is a major distinction between this and, for example, the dot-com bubble. Fiber optic cables have metal and glass basically. They have a useful life of decades. So, you lay it in the ground. You go bust maybe because the profit didn’t materialize right away, but the profit on the cable is still going to materialize eventually. The fiber optic cable is still good.
That’s not the case with these chips. They’re going to be obsolete in a few years at best. And at worst, even if you do use them to full capacity, silicon burns out a lot more quickly than fiber and steel do.
TEDDY DOWNEY: But also, I mean, the business plan of all these things depends on the price not going up 50 percent for all these inputs, the electricity. And you’re seeing already huge price increases for all that stuff.
MATTHEW SCHERER: Right. I didn’t realize that helium—until the Iran war came along—helium is a major resource that is necessary for the construction and operation of a lot of AI data centers. And helium, one of the main sources of helium is, I believe, Qatar, that is being choked off because of the Iran war.
So, again, this boom is a lot more fragile than it looks. And again, that’s scary because there’s so much riding on it kind of reaching fruition and turning a profit within a relatively short time.
TEDDY DOWNEY: We’ve gone through a lot of reasons why we’re in an AI bubble. It’s a fragile bubble. I want to talk a little bit about the history of bailouts. You make an important distinction. There’s a difference between an overt and a covert bailout. Can you tell us about that and give us a few examples?
MATTHEW SCHERER: Yeah. So, the distinction, as far as I know—I always give a hat tip to the intellectual godmothers, in this case, of an idea, and that would be Cheryl Block, a law professor who I believe was at Washington University for a long time. And she wrote this in 1991 in the aftermath of the savings and loan bailouts.
So, overt bailouts are—the best example of that is the bailouts that were given to Wall Street in the aftermath of the 2008 financial crisis. That was a Congressional act that everybody knew was explicitly designed to bail out a particular industry. There was no mistaking it. Was the act called “ A Bailout for Wall Street? No. But everybody understood during the process that was the goal of the legislation. And usually, those sorts of bailouts, they come in the form of direct assistance of some kind, whether it’s a loan or simple cash infusions or loan guarantees. And again, there’s no mistaking that that’s what they’re there to do.
The true costs of them are still not always apparent. You don’t know how much of a loan you’ll get paid back. You don’t know how much you’ll be on the hook for when you guarantee a risky loan. So, just because it’s overt doesn’t mean that the true costs are evident. We learned that in the savings and loan bailouts and in the 2008 bailouts.
But you can also have bailouts that even their purpose is not obvious or evident. And the example of that—that Cheryl Block pointed to, that I think is the best example—is basically when you allow a large monopolistic company to absorb a struggling smaller one. Antitrust considerations would typically say you shouldn’t do that. But there’s a doctrine that’s developed over the years that basically says if a company is in dire financial straits, we’ll let an industry giant buy it and take it over in order to make sure that the rot doesn’t spread, essentially. And it’s an easy way out for regulators and for policymakers to allow that to happen. But it does damage to consumer choice in the long run.
And an example of that dynamic at play is the airline industry. The airlines got an explicit bailout in 2001 in the aftermath of the September 11th attacks. When that proved not really sufficient to keep the airlines profitable in the long run, regulators essentially looked the other way as they started merging and consolidating with each other.
And so, now what we have is fewer airlines with worse service and higher prices and lower quality and fewer choices for consumers. And that was a covert bailout of the airline industry. It harmed consumers and essentially allowed the executives and shareholders of these companies to enrich themselves at our expense. And nobody thinks of that as a bailout, allowing those mergers to happen. But it was a bailout. We suspended the usual rules—under which our economy is supposed to operate—in order to allow an enterprise that otherwise would have failed and entered bankruptcy to do so.
So, that’s an example of a covert bailout. Another one that I cite in my article is the change in rules to allow private equity companies last year to have 401k investors invest in their funds. And I think that now that we’re seeing those private equity and credit funds, a lot of them struggling this year—so soon after that rule change was announced—is a strong sign that that wasn’t actually about giving workers and retirees investment opportunities in these funds, but rather to give them a source of new funding to pay off their old investors.
TEDDY DOWNEY: This is so interesting. Because a covert bailout can be changing the rules to let people merge when they otherwise wouldn’t allow it. Or changing the rules for how your business operates. And that can be kind of subtle. Sometimes it’s a little bit more obvious, a big tax credit or some other kind of protectionist policy or what have you.
But you’re saying even a slight regulatory change that private equity is running out of investors. So, hey, let’s open it up to 401k investors and pension investors. And that way, all of a sudden, you’re getting a lot more money for private equity that they otherwise, weren’t allowed to touch. You have all sorts of rules. A credit investor would have you preventing that typically. But then all of a sudden, now you have access to all that if you’re a private equity firm.
And then on top of that, private equity and private credit are also behind a lot of these, I think, kind of shady financing vehicles for the AI boom. And so, that’s even, you could argue, almost a covert bailout for this AI bubble. It’s potentially extending, allowing this bubble to continue. Where otherwise, it might not be able to. I thought that was a super interesting point. Am I overstating the case there?
MATTHEW SCHERER: No, you are absolutely right. I think that, number one, it was a subtle rule change, but it is a big rule change. Historically, there has been an iron wall that has said we are not going to allow retirement accounts that people are going to use after they are no longer earning employment income to be invested in opaque, shadowy, murky, risky markets that private equity funds and hedge funds operate. That has been a rule and there’s good reasons for that. Most workers do not actively manage their retirement accounts. They invest them in funds and they kind of trust fund managers to do the right thing with them.
Well, we know from past experience that fund managers are very susceptible to pressure from their bosses to not only generate higher short-term returns, but also to maintain good relationships with the funds and firms that generate high returns. And to be blunt, retirees are not their top priority. So, given the choice between giving a wealthy client access to the juiciest funds and keeping the riskiest funds away from them, and doing the same with retirees and workers, what we’re going to see—I have, just from knowing how similar dynamics have played out in the past—workers and retirees, I am very deeply concerned, are going to be the ones who end up stuck with the riskiest investments in their portfolios.
And again, we’re kind of already—the fact that we’re seeing distress in the private equity and credit funds recently is a strong signal of that. What I point to that as kind of the strongest evidence of that is the amount of redemption requests that we’ve seen on these funds in the last few months.
And just to explain what that means usually when you invest in a private equity or credit fund, you’re supposed to keep that money locked in with them for a certain period of time, like five or ten years. But if the investors in those funds start to think that money’s not going to be there anymore, that the investment’s not going to pay off, they might get antsy and say, hey, we want to get our money back. That’s called a redemption. You get your money out of a fund. It’s no different than getting your money out of a bank.
If you think the bank’s going to fail—if enough people do that—if enough people think that a bank’s going to fail, what you end up with is a run on the bank. Like you had with Silicon Valley bank in 2023, like anybody who’s watched “It’s a Wonderful Life,” sees with Jimmy Stewart’s Building and Loan Association. And that’s essentially what’s happened with a lot of these private credit funds.
And in order to keep investors from fleeing, they’ve had to come up with things that are very, very Ponzi schemelike in order to pay off investors. And it’s a little bit too convenient that this massive source of new investment dollars from investors that are not actively managing their funds, namely workers and retirees, suddenly becomes available to them. And they successfully pressure the government to make it available right as those funds are teetering on the edge of collapse.
So, that’s a strong suggestion that this was really more of a bailout for private equity and credit than it was an investment opportunity. And like you said, what it also does—and this is what I fear will happen—workers and retirees will get stuck with these underperforming funds that these private equity and credit fund institutions and firms have been operating for the past few years. And the investors who are now with the cash out of those funds, they’re going to use those investors’ money to invest in the AI boom. That’s what we’ve been seeing.
So, yes, it is a kind of—I wouldn’t call it a backdoor bailout yet of AI because right now the bubble is still inflating. We are not yet in the stage where the bubble is deflating and you see people cashing out of these AI projects. What it’s doing is you are simultaneously bailing out private equity and credit funds from their old bad investments. But in my view, you’re pumping the money into new bad investments.
TEDDY DOWNEY: Right. I think that’s super interesting. We’ve done a lot of reporting on how private equity and private credit funds are buying up insurance companies and then using weak regulators to just push more risk into the assets that these insurance companies are allowed to hold. That’s another kind of backdoor—not necessarily a bailout, but could be kind of, if you think about it. It’s not necessarily a rule change, but they’re sort of gaming the system.
So, I think it’s interesting to think about all of these rule changes that you wouldn’t necessarily equate with a bailout to, is it a little bit of a bailout? And you actually mentioned tax credits and you say that $51 billion in tax breaks in 2025 alone for Meta, Alphabet, Tesla, and Amazon. Again, you’re talking, maybe this is more in the sense of like it’s fueling the boom as opposed to creating a bailout. But it does feel like these incentives, you’ve got to keep an eye on them to make sure they’re not in bailout territory at some level.
MATTHEW SCHERER: And my report has a section where one of the bailout arguments that we should expect to hear—and that we are already hearing from the AI sector—is it’s not a bailout. It’s an “investment.” And the distinction between the two is very blurry, especially when you are kind of at the point where the parabola of the bubble is about to turn downwards. At that point, usually companies are the first to know when they’re in trouble. Not always. But usually, companies know that they are in distress and that they need infusions of cash before regulators and policymakers do.
TEDDY DOWNEY: Well, in the report, you even mentioned you’re already seeing some sort of bailout style requests from the industry. Can you walk us through that? And then what else do you see happening in the future in terms of other arguments about like why they should get bailed out if things get ugly?
MATTHEW SCHERER: Yeah. So, there were a couple of what, in my view, were unmistakable trial balloons from the AI industry late last year for bailouts where Sarah Friar, the Chief Financial Officer of Open AI, basically floated the idea of a federal government backstop for AI infrastructure investments.
And then David Sacks, the White House AI Czar and a venture capitalist, made this somewhat cryptic X post saying that basically the economy is dependent on AI. We can’t afford a downturn. And the implication—the clear implication—from it was that we needed to prop up the AI industry at all costs. To me, those were both unmistakable trial balloons for a bailout. They both walked their comments back after there was pretty intense public backlash to the suggestion.
TEDDY DOWNEY: Not just trial balloons for a bailout, but also indicia that we’re in a bubble, I would add.
MATTHEW SCHERER: Right, exactly. You don’t need a government backstop for these investments if you have a revolutionary technology, that’s going to earn $26 trillion—that you actually believe is going to earn $26 trillion over the next few years. At least that’s my take. Like, if you’re that confident about it, you’re not going to—because the government doesn’t invest in things, at least typically, with no strings attached. You’re going to want to invest your own investors’ money and reap all the profits. I say at one point in it, the AI industry has shown no desire to share their profits or the gains that they hope to get out of AI with the general public before. So, beware of any investment opportunities that they offer now.
But what I think has happened since those couple of trial balloons got floated and shot down is they’ve shifted their rhetoric. They’ve really seized, frankly, on the Iran war. And they are much more openly seeking government funding for direct federal government use of AI infrastructure and construction of AI infrastructure.
And there’s all sorts of ways in which the Pentagon budget, infrastructure spending, chips,—that spending can be repurposed or diverted so that it instead goes to purchase assets that otherwise might be declining in value when the bubble bursts for the AI industry.
So, in my view, they are still very much seeking bailouts. But they have become—it’s the distinction between last fall, what they were floating was really more of an overt bailout and what they’re now doing is seeking covert bailouts that they’re hoping will fly under the radar.
TEDDY DOWNEY: And one of the things you mentioned is that AI is going to say, well, we’re important to national security, to your point about the Iran war. We’re important to beating China. And they sort of create this almost mythical battle that’s happening over chat bots. It’s very confusing to me.
You point out some of the inconsistency in the paper about like, well, is this really important from a national security standpoint? Why did you conclude in the paper that it’s not? And also, that’s one reason why policymakers should reject these covert bailouts. But any other reasons why they should as well?
MATTHEW SCHERER: Yeah, I would say, I’m not the one who concluded that they are not a boon to national security. It was researchers from the U.S. Army War College and the Navy that concluded that these things are potentially harmful to military readiness and preparedness in our national security. The Navy actually, a couple of years ago, effectively banned the use of generative AI in a lot of settings, basically anything related to combat use of these technologies. And professors from the U.S. Army War College really, on the eve of the Iran war, published a study saying that using these technologies gives our military a glass jaw. Because commanders have a tendency to over-rely on the outputs of AI generated systems, to not think critically about their accuracy or verify them. And that they end up, as a result, kind of being disengaged from the command and control process that needs to be really tight—and frankly, deterministic, in a way that these statistical/probabilistic chatbots are not—in order to make good decisions.
Now, one thing that I do point out is that—and there is another piece that I came across—that I don’t remember where it’s from or like where I read it, but maybe one of the readers will be able to drop it in the chat—about how we should really stop using the term AI in military settings. And maybe we should stop using it completely because it’s being used to refer to such a huge range of technologies.
Really, when most people say AI today, they’re thinking about generative AI. They’re thinking about large language models. But there are other applications of AI that really do have strong military applications. Basically, things like data analytics, sensors that are networked together that allow for better targeting. But these are not the systems that we are spending $7 trillion on and that the AI industry is betting our economy on.
And so, I think that there’s some word games that happens where they say like, hey, AI is important to our national security. And no doubt, some things that fall under the broad umbrella of AI are going to be important in military and intelligence settings going forward.
The question is whether those are the things that $7 trillion is being spent by the private sector and building out. And whether the government, when it is investing money in AI—are they investing are they investing in these kinds of technologies that are more deterministic and more reliable? Or are they investing them in essentially buying up new data centers and training bigger and better generative AI models, which are not reliable? And which again, right up until Pete Hegseth and the military decided that they were going to lean heavily into AI, right up until that moment, the military experts were saying, “this is a very risky technology to incorporate into national security applications.”
TEDDY DOWNEY: Well, we saw one of the examples of that. In your paper, you mentioned how there’s been some reporting that when the U.S. bombed that school in Iran, that it was using gen AI targeting or something that affects. So, the hallucinations or the errors are just enough that you can’t really rely on this for sensitive decisionmaking, as you point out.
I want to open the floor for questions listener questions. If you have a question, please put it in the chat. We have a few. I want to get to them momentarily. We could spend all day talking about all the different arguments that these companies are going to make to try to get bailed out and why they should be rejected.
Can you walk us through quickly though why it’s okay to reject these bailout arguments when they come because these companies are not systemically important, like a financial institution. I think that’s an important thing to touch on before we get into the what should we do instead segment?
MATTHEW SCHERER: For sure. So, two things. Number one, even if an institution is systemically important, that doesn’t mean a bailout’s a good idea. That’s a disclaimer I give at the beginning. There are better ways to deal with the rise of systemic risk than giving taxpayer funds or suspending the usual rules that govern our economy in a way that essentially allows a company that made bad investment decisions to get out from under the obligations that they’ve taken on.
But even if that is the case, even if there are some systemically important financial institutions where a bailout in some hypothetical circumstance might make sense—the tech industry, tech companies, are not banks. They do not hold the deposits of ordinary people that people use to pay their bills. They do not lend money to small businesses that need to make payroll. They do not hold commercial paper that they are constantly using as collateral in order to keep the plumbing of the financial system working.
The reason that the bailout happened the way it did in 2008—and the book “Too Big to Fail” by Andrew Ross Sorkin and a far too short movie that I hope somebody makes into a miniseries someday based on it. But there’s a great scene in it where Ben Bernanke, played in the movie by Paul Giamatti, says like, “Look, when people stop lending money to each other, when they get too scared and they stop lending money to each other, the economy just seizes up.” That happens with banks. It does not happen with tech companies.
So, no matter how big a tech company is, there’s no reason that it can’t go through the usual Chapter 11 bankruptcy process if it goes belly up. That’s true, certainly, if it’s OpenAI or Anthropic or CoreWeave it. But frankly, it’s even true if it’s Apple or Nvidia or Alphabet. These are all companies that have assets that are valuable, but they are not assets that are essential to the functioning of the financial system.
So, that’s the kind of big distinction. And the reason that, yes, these are big companies, but the idea that they are too big to fail in the sense that led regulators to bail out the banks in 2008—that doesn’t really apply here.
TEDDY DOWNEY: Once that really jumped out to me—and you have a couple of examples of this, but I just want to name one. Nvidia and Alphabet have fewer employees, 230,000, than GM in 2008, 242,000. It would just be funny to also map their market cap versus their employee number which would make it even a starker difference.
What should we do? What should we be doing? I think this is very compelling. I too am a skeptic. We did a whole call on cognitive surrender that you mentioned, where it’s dangerous because the military decision-makers defer to the algorithm. I just have so many. We’ve done enough calls that I’m pretty—I find your evidence and others compelling that we’re in a bubble, that it’s probably going to pop. Although, the timing is unclear. But what should we do when the AI bubble pops, when the bubble bursts, to have a conversation and policy discussion and decisions about a bailout that actually makes sense.
MATTHEW SCHERER: Right. And this is actually a good way to also answer the question that I saw in the chat from Christian about how do we weigh the downsides of recession resulting from a potential AI downturn?
Because one thing that I kind of anticipated is that it’s all well and good to say that when a crash seems imminent, that you don’t bail out the tech companies, but does that mean that you’re just going to let the economy chaotically collapse? No.
And I think that the way that you deal with that—and I found this great licensed image that I use in the report of dominoes falling and you somebody is putting down their hand and keeping some of the dominoes from knocking other dominoes over—that to me is kind of like the visual metaphor that you go for. You let the companies fail safely that are not systemically important. And again, I don’t think any tech company is systemically important such that it can’t go through the usual bankruptcy process. If they have valuable assets, they can be sold off in bankruptcy. If they don’t have valuable assets, you use what’s left and you pay off their creditors. That’s what’s supposed to happen when a company goes bust. There’s no reason it can’t happen with tech companies.
Number two, you need to make sure that the things essential to the functioning of the real economy continue to operate. Now, there is definitely a high risk of this because so many private credit institutions are exposed to the AI bubble. If those start going under, it could easily spread to the full financial system.
Well, at that point, that’s a good justification using the Dodd-Frank resolution process for systemically important financial institutions. If those prove to be inadequate—or if, as the time gets closer, we think that the Dodd-Frank resolution process is not good enough, or that too many key smaller institutions don’t qualify for it—which there’s a smaller number of institutions now that aren’t considered systemically important due to recent rule changes. Then there’s other ideas.
One of which I’ll point folks to is the idea of speed bankruptcy or Super Chapter 11, which Joseph Stiglitz and a couple of other economists—actually from across the political spectrum, including some libertarian economists—have gotten behind it. Basically, what happens in speed bankruptcy is you just hit the fast forward button on what usually happens in the bankruptcy process. And you say, okay, shareholders in a company are gone. The unsecured creditors or bondholders are the new owners of the company. Snap of your fingers. It happens overnight. There’s no time for people to panic.
And if that’s still not enough, then, for the financial system, there are other solutions that I’ll be exploring in another report later this summer. But really the idea should be if you make bad investments, you shouldn’t get out from under those. What you should do instead is you should make sure that the banks themselves continue operating, that they continue providing the liquidity that companies need to make payroll, that people need to pay their bills, that you save any money that might otherwise be spent on a bailout and ensure that kind of the basic functions of the continue operating normally. Not the functions that are just personally important to the bankers themselves but the ones that actually ensure that, again, the normal small businesses and large businesses can continue paying their employees, providing goods and services to the public. And there’s a large literature on this out there that I’ll be delving into more deeply in another report later this summer.
TEDDY DOWNEY: What about money for the people who are left without jobs or otherwise face dire financial circumstance? Maybe your 401k is just bottomed out or what have you. What’s the equivalent—if you look at 2008, we bailed out the banks. We found the runway for them. But little to nothing was done for the average homeowner who was defrauded right in the first place, a lot of them. Is there an equivalent harmed citizen who you think should get direct support or should be the focus of any bailout as opposed to Big Tech?
MATTHEW SCHERER: I think that it’s going to be much more tangential—or not tangential. It’s going to be people a couple of layers removed from the core of the bubble. Because, with housing, part of what made that so visceral was that it was about the place people live. It was about your home, your shelter. For AI, the biggest risk is not direct—you know, other than the fact that, yes, there’s more people invested directly in the stock market than ever before—but most ordinary people their finances will not fall apart, even if the value of Nvidia and Alphabet go to zero, which they won’t in all likelihood.
The real risk comes from all of the other ways in which the AI boom has gotten tied up with the real economy, with so many insurers and pension funds deeply invested in private credit funds and in particular seeking the juicy returns that they can get from AI infrastructure projects.
And also, people are already seeing utility rate increases because of data centers. But actually, that could get a lot worse if data centers start going offline and all of a sudden, all of this new infrastructure that utilities have been building to power data centers, no longer have these huge corporate customers to pay off those infrastructure debts. Then what’s going to happen is ordinary rate payers, ordinary households and small businesses, will be left holding the bag in order to pay off those debts. And if utilities aren’t able to raise rates enough to pay off those debts, then they’re going to go bankrupt.
That’s actually like an example of the sort of thing where, whether you want to do it as a bailout or whether you say we’re going to let the utilities go bankrupt and convert them into co-ops, which is another model of operating utility. But the point is that there’s going to be all of these ways—there’s all these paths through which the bursting of the AI bubble can trickle down to the real economy and to ordinary people. And that is where the financial resources and the policy responses need to be focused.
TEDDY DOWNEY: Let’s just take a couple more listener questions. If you can just bear with me on that, then we’ll let you go. Actually, I think we’ve already answered this. How should our financial stability tools be used to address a potential downturn? We already talked about that last one.
The covert bailout discussion is so interesting. There is currently a huge push for private credit to get their hands on 401k funds. And there’s a deal department of labor notice of proposed rulemaking gathering comments right now on that. And there’s a new trade group that is hardcore lobbying for the effort, the Pinpoint Policy Institute. Sounds very shady, just calling yourself the Pinpoint Policy Institute. The U.S. Department of Labor proposes a landmark rule to democratize access to alternative investments in 401k plans.
This language is just like kind of Orwellian—democratize access pinpoint policy Institute. Do you see a lot of that in you, do you, does that concern you this kind of, language, PR, all the money in the lobbying and public relations that you see from the AI bubble, the tech, Big Tech companies, does that concern you that it will be hard to, for your paper to break through and get the dialogue that we would need to get a good decision here?
MATTHEW SCHERER: It does. And it terrifies me the extent to which it’s kind of an uphill climb to get this issue as high on policymakers’ radar as I think it needs to be. Like, I think that this is—I’m not going to mince words. I think that this has the potential, with all of the other pain points that are building up in the economy at the same time between private credit, between the massive amounts of leverage that are built up in margin accounts, between the amount of wealth that people have tied to crypto, which could easily go to zero in a deep enough downturn—these are the makings of a major economic crash. And it’s not being treated as a high priority right now.
And I do think that it is because of the influence of industry on this. And quite frankly, the reason that I took my current role at OMI is that in my last role, I was doing policy advocacy on workers’ rights in technology with the Center for Democracy & Technology. And I repeatedly ran into these kinds of astroturfed efforts by the tech industry that frankly misled policymakers into swallowing the tech industry’s narratives on a lot of things. I won’t go too deep into detail on that, but I am deeply worried that so much of the policy world has been taken in by this narrative of, “we are in a transformative moment with AI.” And there is this natural tendency to believe that because AI is—or could be—a transformative technology that therefore any concerns about the economics of a potential bubble or a bust are secondary to preparing for this massive transformation.
And if there’s one thing that I want to leave anybody who is watching this webinar with, it’s that even if you believe in that narrative, even if you believe that in the long run this is a transformative technology, you still could have, quite frankly, a 2008 or 1929 level economic crisis arising out of it, if that investment takes much longer than these companies are saying to pay off.
But I do worry that as long as that narrative continues to kind of dominate tech policy discussions, it will be very difficult for this message to break through. But hopefully, events like this and similar conversations will help raise the awareness of these issues on policymakers’ radar.
TEDDY DOWNEY: I want to leave with one more question really quickly. My worry is always that the Fed and Treasury under Trump are just going to get together and come up with the money that they need, a vehicle that they need, to just bail out David Sachs and all his buddies in the AI space. I know you have a slightly different take on that. Can you share that with us?
MATTHEW SCHERER: Well, number one, I think you’re right. I think that they will try to do that. But what we’re talking about, again, like, the AI build out is $7 trillion. We don’t know exactly how much of that is in the form of debt, but it’s north of $1 trillion. It might be as high as $2 or $3 trillion. And that amount is going up rapidly.
That is not something that the Treasury and the Fed can cover using spare change from the budget and printing new money, at least not without having truly distorting, massively distorting, effects. It would cause all sorts of other economic problems and potentially a constitutional crisis, depending on how the money is allocated and dispensed.
So, I think that that sort of move, it will likely happen. But I think that the most that they could do without Congressional action is kick the can down the road by a few months or maybe a year or two. Ultimately, the size of this bubble is such that, in order to prevent massive economic damage when it bursts, one way or another, it’s going to require Congressional action. And that is where the rubber is ultimately going to meet the road.
As much as David Sachs, I’m sure, will try to convince the Trump administration and the Trump administration will try to lean on the Fed to do as much as they can without Congressional allocation of funding and authorization, I think that the scale of the bubble is such that ultimately they’re going to need Congress to step in. And that is where I’m hoping that, with this report in folks’ hands, they will be able to kind of recognize those arguments and rebut them and see them for what they are.
TEDDY DOWNEY: Well, Matthew, this is a truly interesting conversation. I really look forward to following your work and especially as it’s debated in Congress, hopefully ASAP. Thank you so much for doing this.
MATTHEW SCHERER: Thanks very much for having me.
TEDDY DOWNEY: And thanks to everyone for joining the call. I just want to let people know, please check out The Capitol Forum’s podcast series, TCF Investigates. We have other episodes of Second Request, along with Matthew’s talk, and we get new episodes every week. So, please check that out. Thank you so much for joining us today. This concludes the call. Bye-bye.
