Is the AI bubble about to burst?
Tech giants like Alphabet, Amazon, Meta, Microsoft, and Oracle are betting over $4 trillion on data centers, chips, and AI models – but the profits aren’t there yet, and the cash is running out. Michael Roberts, economist and author of The Next Recession blog, joins Zoe Alexandra to break down the reality behind the AI boom – what he calls the biggest economic gamble in US history – and who will pay the price when the house of cards comes down.
Roberts explains how circular financing keeps AI companies afloat, what Chinese open-source models undercutting OpenAI and Anthropic mean for US workers, and why grassroots opposition to data centers is a reasonable response to hyperscale projects that drain power and water while creating few jobs. He warns of a 2008-style recession on steroids, and explains why following the money today matters more than the “shock message” of a coming AI apocalypse.
Read Michael Roberts’ article “AI: The Biggest Bet in U.S. History”: https://thenextrecession.wordpress.com/2026/09/26/ai-the-biggest-economic-bet-in-us-history/
For more deep dives into the global economy, tech monopolies, and the real-world impact of AI – and to watch full episodes of BT Live, join breakthroughnews.org as a member today!
#breakthroughnews
******
Rough transcript follows:
[Speaker 1 (Zoe):]
Is the US economy growing incredibly fast in the last year or is it a bubble that is about to burst? Tech giants like Alphabet, Amazon, Meta, Microsoft and Oracle are expected to pour more than four trillion into data centers, chips, and AI models by 2029. Brookings estimates total data center spending could reach 10.3 trillion by 2032. And that’s more relative to the GDP than railroads, the highway system, or the internet. Up to July, 37 billion has actually been spent on private data centers with most still not operating. In contrast, US private construction spending on everything else that it does, houses, apartment buildings, shopping centers, and so on was about 46 billion dollar below the levels in the year prior in the first seven months just of this year. So that’s a marked difference and a marked contrast. All of this is part of what experts have been referring to as the AI bubble. And so to discuss this, what does it mean for everyday people in the United States and the world, we are joined by Michael Roberts, socialist economist, author of several books on the world economy, and the author of a blog called The Next Recession. Thank you so much, Michael, for joining us today here on BT Live.
[Speaker 2 (Michael Roberts):]
Pleased to be with you, Zoe. An important matter to discuss, I think.
[Speaker 1 (Zoe):]
Extremely important. And your latest article, AI, the biggest bet in US history, was very important. We’re going to link it in the show notes because I highly recommend that everyone read this article. But of course in this article, you start out by saying that AI is the biggest economic bet in US history. As you report in this article, spending is set to pass 1 trillion next year. However, the free cash flow funding is falling below 100 billion. So, explain to us what does that gap mean? How does that impact ordinary people and is this something that we should be worried about?
[Speaker 2 (Michael Roberts):]
Well, it seems that American institutions, the financial institutions, companies, government are absolutely placing a huge bet on AI delivering faster growth for the United States, better wages for everybody, higher profits for companies. In fact, a whole new wonderful world is going to come from this AI, artificial intelligence. And that in order to do that they want to spend huge amounts and are spending huge amounts of money. Both the banks are lending money, private credit institutions are putting money into AI labs and into data centers dotted across the United States and there’s still more to come. This is a massive investment. Will it come off is the question. And also before it even does come off, will the money run out? Because already most of the money that’s been provided for this investment up to now has come from the profits from the big tech media companies who were making lots of profit from previous businesses, cloud businesses, internet businesses. Now they’re running out of how much money they can put in. So they’re now borrowing money. So the debt is rising to cover these AI investments which are taking place. And now within the next month or so, one of the big AI labs called Anthropic is going to launch a share issue where it wants to raise $2 trillion, a massive amount of money which it claims if it gets it will be able to deliver fantastic AI models that companies can use in every sort of part of their business and transform the situation, probably at the expense of a lot of jobs unfortunately for people who are currently working in them.
[Speaker 1 (Zoe):]
Well, I want to touch on something that you illustrate in the article with a slightly complicated graph, but you explain it very well, which is this concept of circular financing, and you explain how AI companies are essentially staying afloat because of this issue that you illustrated of lack of free cash flow funding. And you explain how firms are lending and buying from each other, and it seems that everyone is essentially dependent on each other’s spending. So can you exactly explain how this works and why is this actually a risk instead of a normal business practice?
[Speaker 2 (Michael Roberts):]
Well, it is in a way a normal business practice to lend money to companies in the hope that they will produce sales and revenues and that then they will get their money back. But that’s not the sort of lending that’s going on now. What’s happening is that the big companies in tech are lending money to the artificial intelligence labs as they’re called. The people who are making these models and these models are then claiming they’re making huge amounts of revenue. But the only way they’re making huge amounts of revenue is they’re getting loans from the other tech companies. So for example, Nvidia which makes all the chips for all these AI companies and for these data centers basically invested in all the AI companies and are lending the money to all the companies. So the result of course is that the companies appear to have plenty of cash but they have an obligation to deliver at some point huge amounts of revenues back to Nvidia. In the meantime this money keeps circulating around and we call it circular financing in the most complex way. That graph shows how much connection there is between all these different companies at all different levels not just in the US but right across the world now. And so if one company particularly a big one starts to lose confidence in making those investments or loses revenue and can’t achieve the targets that it has to make then the whole of this web could start to crumble and cause the bubble to burst and we see a massive reduction in the value of all these companies on the stock market and that will produce losses of jobs, losses of income for large numbers of people and we could then slip into a recession in the United States. That’s the risk. It’s so we have one huge bet now on the US economy that AI is going to work. It’s going to deliver huge profits for the companies that have made these loans and it’s going to deliver greater productivity for the economy as a whole. That’s a huge risky bet.
[Speaker 1 (Zoe):]
Well, you know, it’s interesting your piece comes out and we’re discussing it right now in the midst of this whole AI news crisis where we’ve seen several whistleblowers, so to speak, come forward and say, you know, AI could kill us all. AI could end humanity. We are racing towards a point of no return, talking about the possibilities that their AI models are unleashing on the people of the world. And it’s interesting because what you’re outlining in this article is that there’s a very real risk that will come before all of that, which is the economic impact. And so what do you think could be the fallout of any of this starting to collapse, whether it’s the circular financing or some of these investments not returning?
[Speaker 2 (Michael Roberts):]
Yeah. Well, I think an estimate was made by the IMF recently and they reckon that it could reduce the stock market value by up to 25%. It could knock back our national output in the US and in other major economies by four or 5%. If you remember the great recession that some of us remember back in 2008-2009 knocked about 3 or 4% off the national output of major economies for several years. This would be just as bad if not worse if we get into that situation. That’s the risk.
Now I have to say that at the moment stock market investors, financial institutions, AI labs, the big tech companies are completely optimistic. That’s not going to happen. They say they’re going to be able to deliver much better models which companies can use right throughout their businesses, get rid of a lot of costs, and dramatically increase their productivity and therefore sell more goods and services, and it will transform the situation for the US. The chairman of the US Federal Reserve said this is the most important thing. It’s going to keep prices down. It’s going to increase productivity. It’s going to transform the economy. That’s the optimistic message that remains.
Who actually mentioned, Zoe, the pessimistic message is it’s so powerful this AI apparently, it’s going to take over human beings’ jobs, but is actually going to control human beings and even annihilate them. I think that’s a sort of shock message which I think tells you two things: one is that the AI labs want to convince you that this is absolutely powerful and will work, and secondly, if it’s dangerous they’d like to slow down the process a little bit, so they want it both ways at the moment. They don’t want to spend too much more money. On the other hand, they want you to believe that it’s going to deliver. And that’s the big question we have now.
So I could call it a $64,000 question, but what we’re talking about is a four or $5 trillion question, which is if you know that the US economy produces labor and machines produce about $30 trillion a year, you can see that if that happened the loss of $5 trillion every year for the next five years that would be a disaster for the US economy.
[Speaker 1 (Zoe):]
No, that’s duly noted. And you know, let’s talk about data centers a bit because in your piece you wrote about how $37 billion has been spent on private data centers, and notably you included a graph showing that a lot of these are not operational yet. And across the United States and in many parts of the world, the construction of data centers has been met with mass grassroots opposition. Recent polling in the United States shows that not only do progressives and Democrats oppose the construction of data centers in their neighborhoods and their areas, but it’s actually something that’s a bipartisan issue for voters. Even Republican voters have joined in this mass opposition to the construction of data centers. How does that factor play into this? Because again $37 billion, it’s not $4 trillion, but it’s still a significant amount. Will the grassroots opposition to the construction of these centers have an impact on how this is playing out in the economy? Is that—we’ve seen a lot of vilification by these tech billionaires against people who are protesting the data centers saying that they’re anti-American. Break down a bit of this for us.
[Speaker 2 (Michael Roberts):]
Well, there are genuine worries for people who have got data centers near them. Data centers consume huge amounts of water. Why is that? Because there are computers and chips buzzing away in these data centers, increasing dramatically the temperature of what’s going on inside. So, they need to be cooled all the time. So they need a large amount of water to keep cooling the center down so that nothing catches fire or breaks down. At the same time they’re using huge amounts of electricity. They’re sucking in electricity in order to keep these computers going. And this is obviously going to have an effect both on the power situation and the water situation in many parts of the country wherever these data centers have been built. And that’s a perfectly reasonable, genuine grievance that people have: do we need this massive expenditure that’s going on in these centers sucking away power and water for us? And remember, people are getting jobs by constructing them, but once they’re built they don’t really have many people working in them. So they’re not going to increase employment dramatically at all.
Worse than that, Zoe, as you saw in that graph, large numbers of data centers that have been promised to be built haven’t been built. But companies have bought huge amounts of chips from Nvidia and other companies to put in these data centers because these semiconductor chips, that’s what really is going into them. And then they can’t put them in because the centers haven’t been built. So they’re sitting in warehouses just sitting there waiting to be used over the next 5 years maybe, because it takes about 3 to 5 years to complete a data center. So in that situation this is a real risk. Large amounts of money have been spent on computers, on chips, and started to construct data centers. But will people use them? Will companies use the AI models that are being produced to run their businesses and therefore use the data centers as a result? It remains to be seen. If they don’t, then there’s going to be a huge loss not only for companies that have invested in the data centers and the people looking to sell goods to people through the data centers, but also the people living next to the data centers will find that these things have been built for no reason whatsoever because they can’t deliver.
[Speaker 1 (Zoe):]
And you know, another aspect of this whole AI economy risk is the fact that the US is not the only country that is producing AI. It’s not only these US tech giants that are producing cutting-edge AI models, and you note the rise of Chinese open-source models, which are significantly cheaper in terms of the cost invested into creating them, but also the number of data centers etc. Overall, it’s a much cheaper investment in these open-source models, and this is overall driving prices down. And you refer to this as a classic contradiction, because there may be a rise in productivity, but the profits will also be kind of destroyed to be able to pay for this. So talk a bit more about this: what role do these open-source models play in this kind of drive to invest and this AI bubble in the United States?
[Speaker 2 (Michael Roberts):]
Two things. Open source means that basically if you’re a computer expert—not like me or you—you can go into the model and you can work out how it works, and you can get all the programs and algorithms out of it, and you can build your own model or use it for free. The models that the American companies are selling like Anthropic and OpenAI are not open source; you have to pay large amounts of money if you want to get control of that model—it’s proprietary and to use it.
Also, that means that Chinese companies, but not just Chinese companies, but other open-source companies in France and in other parts of Europe can sell their time when you go and prompt an AI model. You buy a token to use something in your business or for your information, and they can sell their tokens at virtually nothing at all because they’re not really charging any money for the source models that they’ve produced. This undercuts the position of OpenAI and Anthropic dramatically. They’ve had to reduce their charges for tokens, as they’re called in this business, for information and algorithms that companies can use in their businesses.
So there are two things here. Companies can set up their own open-source models to use in their companies; they don’t have to use OpenAI or Anthropic maybe, and also they can use the Chinese models or the open-source models very cheaply. So as you correctly say, this may improve the productivity of companies that are buying these things and using them, but it means that the profitability of the existing AI companies and the people who have made all these investments and loans to them may not be delivered as a result.
[Speaker 1 (Zoe):]
And you know, just before we let you go, I wanted to ask you one last question. It feels like we’re kind of on a precipice here, as you’ve laid out in terms of this bubble. What—of course you don’t have a crystal ball, neither do I—what should we be looking for in terms of triggering events which could change the current situation? It feels quite tenuous right now. What should people be paying attention to in terms of reports or things that could change this balance that you describe?
[Speaker 2 (Michael Roberts):]
Well, this boom, this bubble, if we want to call it that, could continue for another year, maybe longer—can’t be sure. So that’s why actually we can’t be sure. But what we should be looking for, for example, is if there’s suddenly a downturn in the demand to build more data centers, or if the share issue that Anthropic is making in November turns out to be a bit of a dud—they don’t raise the money they expect to do so, or if they raise the money then the price of the shares afterwards drops sharply. That would indicate a lack of confidence in the possibility of these AI models delivering what they claim they can deliver over the next few years. And if they don’t start changing the levels of productivity—if companies don’t adopt these AI models, and sometimes that can take a long time before they get round to doing it, putting it through their own companies and using it in an effective way—if that takes a long time, in the meantime all this money is being spent and it’s not getting a return, we will begin to see a fall in the profitability prospects for those companies and a withdrawal by the investors about making more investments. We see that beginning to happen. That tells you that the turn towards a bust from a boom in the AI stock market and elsewhere is beginning to happen. So it really depends on all of us, all the workers, company owners—not us, but—to embrace this for it not to be a boom-and-bust.
So at the moment they are embracing it. Don’t worry there, if you go into the mainstream media, it’s going to be fantastic. And I hope they’re right in some ways. We don’t want massive numbers of jobs to be lost, we don’t want a huge decline in output or people’s incomes because of this bust. But as we know in the history of capitalism, in the history of stock markets, that’s what happens quite often. And the example at the very beginning of the show when you showed the railway mania that happened in the 19th century, that collapsed before we got the railways; the dotcom bubble collapsed in the early 2000s, then we got the internet. This time we could have a huge AI bubble that collapses. We may still be left with artificial intelligence machines afterwards that we can use, but it’s going to be at a cost to us all, not only in the US, but across the world.
[Speaker 1 (Zoe):]
Well, Michael Roberts, thank you so much for joining us here on BT Live. Definitely recommend to anyone who’s watching, please check out his piece, follow his blog; you will thank us later. So, Michael, thanks for being here.
[Speaker 2 (Michael Roberts):]
Thanks for having me.
Login
