Por Dean Baker
First off, here are the latest numbers:
It appears that the bulk of Nvidia’s chips were being sold to companies that were being subsidized by it through one channel or another. Nvidia bought equity stakes in several neoclouds and AI companies; it provided loans and also gave loan guarantees on the purchase of its chips. This financing hugely boosted its sales in comparison with a situation where it had maintained an arms-length relationship with its customers.
This system of dodgy financing also makes Nvidia’s illicit sales to sanctioned Chinese companies, or their fronts, far more important. Some estimates put these sales as high as 20 percent of Nvidia’s total sales. Nvidia puts the figure far lower. While we may never know the true number, given the dubious financing needed to support other sales, the sales to China, which were presumably full-price sales, carried disproportionate importance for the company’s profitability.
It is not clear what Nvidia or the AI industry will look like post-Huang. Presumably, its sales to China and Chinese fronts are brought to an end. It’s likely that its financing arrangements get much closer scrutiny under a Huang successor, which also means lower sales. This prospect has already been a huge hit to Nvidia’s stock price, with its market capitalization down by close to $2 trillion, a drop of 30 percent from its October peak.
The loss of generous Nvidia support is also likely to be a big hit to the neoclouds and AI companies that were dependent on it. The stock of the largest neocloud, CoreWeave, which was a heavy recipient of Nvidia support, rose slightly last week, but is still down more than 30 percent from its level a month earlier. It seems virtually certain that the loss of the massive financing provided by Nvidia will be a big hit to the rest of the industry.
Interestingly, the hyperscalers seem to be going along as if nothing had happened. This would fit a story where the blowup means they are less likely to carry through with plans to borrow trillions over the next three years to finance a massive data center buildout.
There was a very useful exchange of thoughts on the bubble on Paul Krugman’s Substack, with Jared Bernstein and Jason Furman weighing in on opposite sides of the issue. Jared largely agrees with me that we are looking at a serious downturn when the AI bubble deflates. Jason sees a modest hit to growth, if in fact there is a bubble to deflate.
There is some difference on the size of the hit from lost investment if the bubble bursts. Jason puts it at just 0.3 percentage points of GDP, with the view that the vast majority of the investment is made up of imported chips and other items. That seems low to me, but the bigger difference is over the potential hit to consumption from a drop in stock prices. He covers a range of possibilities, but it seems his central number is $10 trillion. With a wealth effect of 3 cents on the dollar, that implies a loss of annual consumption of $300 billion, or roughly 1.0 percentage point of GDP.
Like Jared, I see a far bigger effect. To my mind, the irrational exuberance of the AI bubble has infected the entire market. By Robert Shiller’s measure, the current price-to-earnings (PE) ratio for the market as a whole is 40, close to the peak of the tech bubble in 2000. In this respect, it is worth noting that even the stock of companies that had little to do with tech also crashed in the period from 2000 to 2002. For example, the stock of GM fell by more than 60 percent, and the stock of McDonald’s fell by almost 70 percent.
Jason rejects this logic by pointing out that Shiller’s measure is backward-looking. Since after-tax profits have risen sharply in recent years to record highs, the current price-to-earnings ratio is far lower. That does not provide much assurance in my view.
The total value of the market is close to $80 trillion (maybe a bit lower today). After-tax corporate profits last year were a bit under $3.4 trillion. To just get down to a PE ratio of 20, profits would have to rise by $600 billion, or 18.0 percent. This would imply a large reduction in wages and/or a further reduction in corporate taxes. I would like to think either scenario is implausible. (FWIW, the Congressional Budget Office projects that profits will fall from current peaks.)
Also, a 20-to-1 ratio is still high by historic standards. A 20-to-1 ratio will only allow for a real return on stocks of 5 percent annually, roughly 2.0 percentage points above the return on inflation-indexed Treasury bonds. Historically, the return has been close to 4.0 percentage points above the return on Treasury bonds. That return would imply a PE of 14.3. Getting there without a plunge in stock prices would require after-tax profits to rise to $5.6 trillion, more than 60 percent above their 2025 level. The redistribution to profits in that case would be $2.2 trillion, which comes to almost $13K per worker. That is a much worse story than a market crash in my view.
Some of us might hope that things go the other way. Then we again tax corporate profits at a reasonable rate, and workers can restore their share of income at least part of the way back to where it was in 2000. In this case, the market would look even more overvalued. Remember, the stock market is not your friend.
Will AI Make Capital Unemployed?
I saw or heard an interesting comment last week that related to the issue of a possible redistribution back from capital to labor. (Sorry, I can’t remember the source.) The story was that Microsoft was worried that AI could allow for the creation of cheap new software that would allow people to avoid buying its incredibly profitable suite of products like Word and Excel.
I have no idea of the seriousness of this risk, but it does raise an intriguing possibility. As noted, there has been a massive redistribution from wages to profits over the last quarter century. A good part of this story is the huge profits earned by companies with substantial monopoly power, like what Microsoft has with its operating system and its set of personal and business software.
The idea that software developed by AI might be able to substantially erode this monopoly power is certainly intriguing. Maybe we can tell Claude: “Develop a word processing system that is completely compatible with Microsoft Word, uses near-identical commands, and gets around all of its patents and copyrights.” If Claude pulls it off, it can cost Microsoft tens of billions a year in profits. (Of course, tons of lawyers will be employed in the resulting lawsuits over patents and copyrights.)
Anyhow, we have yet to see much evidence of the mass unemployment of workers from AI. It could still come, but it’s not there yet. It would be pretty damn great if AI could be used to erode monopoly power in key areas to bring capital shares back to earth. (We’ll get Mark Zuckerberg and Meta with Section 230 reform.)
This first ran on Dean Baker’s AI Bubble Monitor blog.
Dean Baker is the senior economist at the Center for Economic and Policy Research in Washington, DC.
The fallout from Nvidia CEO Jensen Huang’s affairs is continuing. While the stories of circular financing were widely known, it seems it went far further than most of us could have imagined. The Wall Street Journal had an excellent piece that dug into some of the details.
It appears that the bulk of Nvidia’s chips were being sold to companies that were being subsidized by it through one channel or another. Nvidia bought equity stakes in several neoclouds and AI companies; it provided loans and also gave loan guarantees on the purchase of its chips. This financing hugely boosted its sales in comparison with a situation where it had maintained an arms-length relationship with its customers.
This system of dodgy financing also makes Nvidia’s illicit sales to sanctioned Chinese companies, or their fronts, far more important. Some estimates put these sales as high as 20 percent of Nvidia’s total sales. Nvidia puts the figure far lower. While we may never know the true number, given the dubious financing needed to support other sales, the sales to China, which were presumably full-price sales, carried disproportionate importance for the company’s profitability.
It is not clear what Nvidia or the AI industry will look like post-Huang. Presumably, its sales to China and Chinese fronts are brought to an end. It’s likely that its financing arrangements get much closer scrutiny under a Huang successor, which also means lower sales. This prospect has already been a huge hit to Nvidia’s stock price, with its market capitalization down by close to $2 trillion, a drop of 30 percent from its October peak.
The loss of generous Nvidia support is also likely to be a big hit to the neoclouds and AI companies that were dependent on it. The stock of the largest neocloud, CoreWeave, which was a heavy recipient of Nvidia support, rose slightly last week, but is still down more than 30 percent from its level a month earlier. It seems virtually certain that the loss of the massive financing provided by Nvidia will be a big hit to the rest of the industry.
Interestingly, the hyperscalers seem to be going along as if nothing had happened. This would fit a story where the blowup means they are less likely to carry through with plans to borrow trillions over the next three years to finance a massive data center buildout.
There was a very useful exchange of thoughts on the bubble on Paul Krugman’s Substack, with Jared Bernstein and Jason Furman weighing in on opposite sides of the issue. Jared largely agrees with me that we are looking at a serious downturn when the AI bubble deflates. Jason sees a modest hit to growth, if in fact there is a bubble to deflate.
There is some difference on the size of the hit from lost investment if the bubble bursts. Jason puts it at just 0.3 percentage points of GDP, with the view that the vast majority of the investment is made up of imported chips and other items. That seems low to me, but the bigger difference is over the potential hit to consumption from a drop in stock prices. He covers a range of possibilities, but it seems his central number is $10 trillion. With a wealth effect of 3 cents on the dollar, that implies a loss of annual consumption of $300 billion, or roughly 1.0 percentage point of GDP.
Like Jared, I see a far bigger effect. To my mind, the irrational exuberance of the AI bubble has infected the entire market. By Robert Shiller’s measure, the current price-to-earnings (PE) ratio for the market as a whole is 40, close to the peak of the tech bubble in 2000. In this respect, it is worth noting that even the stock of companies that had little to do with tech also crashed in the period from 2000 to 2002. For example, the stock of GM fell by more than 60 percent, and the stock of McDonald’s fell by almost 70 percent.
Jason rejects this logic by pointing out that Shiller’s measure is backward-looking. Since after-tax profits have risen sharply in recent years to record highs, the current price-to-earnings ratio is far lower. That does not provide much assurance in my view.
The total value of the market is close to $80 trillion (maybe a bit lower today). After-tax corporate profits last year were a bit under $3.4 trillion. To just get down to a PE ratio of 20, profits would have to rise by $600 billion, or 18.0 percent. This would imply a large reduction in wages and/or a further reduction in corporate taxes. I would like to think either scenario is implausible. (FWIW, the Congressional Budget Office projects that profits will fall from current peaks.)
Also, a 20-to-1 ratio is still high by historic standards. A 20-to-1 ratio will only allow for a real return on stocks of 5 percent annually, roughly 2.0 percentage points above the return on inflation-indexed Treasury bonds. Historically, the return has been close to 4.0 percentage points above the return on Treasury bonds. That return would imply a PE of 14.3. Getting there without a plunge in stock prices would require after-tax profits to rise to $5.6 trillion, more than 60 percent above their 2025 level. The redistribution to profits in that case would be $2.2 trillion, which comes to almost $13K per worker. That is a much worse story than a market crash in my view.
Some of us might hope that things go the other way. Then we again tax corporate profits at a reasonable rate, and workers can restore their share of income at least part of the way back to where it was in 2000. In this case, the market would look even more overvalued. Remember, the stock market is not your friend.
Will AI Make Capital Unemployed?
I saw or heard an interesting comment last week that related to the issue of a possible redistribution back from capital to labor. (Sorry, I can’t remember the source.) The story was that Microsoft was worried that AI could allow for the creation of cheap new software that would allow people to avoid buying its incredibly profitable suite of products like Word and Excel.
I have no idea of the seriousness of this risk, but it does raise an intriguing possibility. As noted, there has been a massive redistribution from wages to profits over the last quarter century. A good part of this story is the huge profits earned by companies with substantial monopoly power, like what Microsoft has with its operating system and its set of personal and business software.
The idea that software developed by AI might be able to substantially erode this monopoly power is certainly intriguing. Maybe we can tell Claude: “Develop a word processing system that is completely compatible with Microsoft Word, uses near-identical commands, and gets around all of its patents and copyrights.” If Claude pulls it off, it can cost Microsoft tens of billions a year in profits. (Of course, tons of lawyers will be employed in the resulting lawsuits over patents and copyrights.)
Anyhow, we have yet to see much evidence of the mass unemployment of workers from AI. It could still come, but it’s not there yet. It would be pretty damn great if AI could be used to erode monopoly power in key areas to bring capital shares back to earth. (We’ll get Mark Zuckerberg and Meta with Section 230 reform.)
This first ran on Dean Baker’s AI Bubble Monitor blog.
Dean Baker is the senior economist at the Center for Economic and Policy Research in Washington, DC.

Nenhum comentário:
Postar um comentário