What kind of crisis would a bursting AI bubble become?

 

In this post I’m not going to speculate about whether the current boom in AI infrastructure (mainly data centres, mainly in the US) is a bubble or not, but rather ask what would happen if it was and, as bubbles do, it burst. [1] What kind of impact might this have on the world economy? There are three main elements that might be involved in any bubble bursting: a stock market collapse, a collapse in real investment, and financial sector problems.

A stock market collapse is perhaps the least interesting, and least worrying. The current boom in the US stock market is very much led by a small number of tech companies. Now my Google AI overview says that a “stock market collapse centered on the artificial intelligence (AI) sector would likely be highly deflationary, primarily by triggering a sharp recession.” It’s possible, of course, but the key point is that the transmission mechanism between a stock market collapse and a recession is a fall in aggregate demand, and specifically a fall in consumer confidence. One thing we know for sure is that policy can influence aggregate demand. [2]

The most immediate impact of a stock market collapse would be a fall in short term interest rates, as the central bank attempts to avoid a recession. A decline in interest rates will not reverse the fall in the stock market, but it may well offset the impact of the stock market decline on consumption and therefore aggregate demand. Indeed that is exactly what the central bank will try and do.

It is possible, of course, that central banks, and particularly the US Fed, will run out of ammunition because interest rates hit their lower bound, just as they did after the Global Financial Crisis. But then a fiscal stimulus can take its place, and at least in the US my guess would be that Trump will not hesitate to pull that lever for purely short term political reasons. That in turn will make US fiscal sustainability even worse, but I doubt whether that will lead to fiscal stimulus having a insignificant or contractionary effect.

A second element of a bursting bubble might be a sudden slowdown in the amount of investment in new data centres. Less new buildings will be built, and less chips to put in them will be bought. That will have a direct negative impact on aggregate demand, mainly in the US. My own guess would be that we are unlikely to see a sudden stop in such investment, for two reasons. First, AI will still be used and will need testing and developing, and the chips required to do that don’t last very long, so will need replacing. A more likely outcome is that investment falls back as some of the companies bow out of the AI race.

Once again, this is a hit to aggregate demand in the short run, so everything noted above about a stock market collapse also applies. Short term interest rates will fall to at least offset the impact on employment of the investment decline, although perhaps not to offset the impact on GDP.

Both scenarios indicate a key point that is often neglected when talking about the consequences of a bursting bubble. If the bubble bursting hits aggregate demand, then policymakers can replace that demand using monetary and fiscal policy. As long as policymakers are prepared to use these instruments, the impact of shocks that just reduce aggregate demand is not nearly as significant as is often portrayed. Keynes taught us that we could stop demand driven recessions, and he was right.

The wildcard in all this is the financial sector. For some companies the initial expansion in AI investment was financed out of these companies’ huge cash reserves, and it is only more recently that investment has been funded by borrowing. Once the investment is financed by borrowing, the possibility that companies may not be able to pay back (or service) that borrowing arises. If default happens, then the financial sector is hit.

The Global Financial Crisis (GFC) involved three elements, all of which were necessary to produce it. The first, and perhaps the least important, was the collapse of the US sub-prime mortgage market. Why do I say least important? Because shocks to the financial sector are bound to occur from time to time. The financial sector should be robust to such shocks. The second element was how interconnected Western banks had become, so problems in a US market hit banks not just in the US, but also the UK and Europe. (Those that forecast a crisis because the UK private sector was borrowing too much were not spot on: the big UK banks got into difficulty because of the exposure to US defaults.) That was why a US centred crisis became global. The third, and the key, problem before the GFC, was that the banking sector in most Western countries had made itself extremely fragile to such shocks.

The crucial concept here is leverage: the ratio of loans to the capital a financial company has. Here is this data for US banks (source).

G-SIBs are ‘global systemically important banks’. You can see how, before the GFC, banks reduced the amount of capital they had relative to loans, which meant that they became much more vulnerable to shocks. That has been more than reversed since the crisis (largely because of tougher rules imposed by governments), so banks at least should be more resilient to a financial shock such as defaults on loans for AI investments.

It would be nice to conclude that the financial system should now be able to weather an AI shock, but that would be a dangerous assumption for two reasons. First, while banks are more resilient now than they were in 2007, they still will be vulnerable if the shock to hit them is very large. As I argued in various posts, capital requirements for banks should be much higher than they currently are, because there is little justification for the state to implicitly provide an insurance policy to banks against very large financial shocks. Second, the financial system remains both extremely complex and interconnected, so it is possible that some non-bank institutions might fail as a result of a sufficiently big AI shock, and that will intensify any impact on banks. As Paul Kedrosky outlines here, there are elements of the GFC collapse that are being repeated with the AI boom. See also this.

Note also that instances of default and financial concern need not be caused by a complete collapse of the AI bubble. Perhaps more likely is that some firms succeed in generating revenues to pay back loans (through AI or their existing activities) while others do not. What we just do not know for certain is whether this more limited AI financial shock can be absorbed by the financial sector as a whole, or not,

Here I want to make a more general point about financial crises and macroeconomic forecasts. It is extremely unlikely that macroeconomic forecasts will ever anticipate a financial crisis, because to anticipate a financial crisis involves deep knowledge of a very opaque and rapidly changing financial system that very few people have. Before the GFC, some well known economists did know about the increase in bank leverage and did try and justify it with invalid arguments, and at least one other economist publicly disagreed. As I argue here, to suggest that economics as a whole is somehow discredited as a result of the GFC is nonsense. Equally it is foolish to believe that macroeconomists like myself can tell you whether bursting of the AI bubble will precipitate a similar crisis.

What is well known and is a legitimate issue of concern is that policymakers, at least in the US and UK, are currently focused on reducing rather than increasing banking sector capital requirements. The reasons why the current administration in the US wants to do this are obvious and not worth discussing. Rather more surprising, as David Aikman - the new NIESR director - notes in a recent blog, is that the Bank of England's Fiscal Policy Committee also thinks the appropriate capital requirement for UK banks has fallen. Aikman starts his blog with "Did they get the sign wrong?" Readers can judge themselves by reading the latest FPC report, about whether it squares its opening statement "Risks to financial stability have increased during 2025" with this recommendation at the end, and whether pressure from the Chancellor that the report discusses in between is material or not.

None of this suggests that we should be sanguine about the consequence of an AI bubble bursting. But nor is it the case that we should be fatalistic about the consequences of such an event. Policy can do a great deal to moderate the impact of a burst AI bubble. While fiscal policy failed to take such action in a consistent way after the GFC, it did react in a much more appropriate manner during the pandemic, and in the UK at least may have provided too much fiscal support. The biggest risk, but also perhaps the least likely, is that a bursting bubble creates insolvencies in parts of the financial system.

[1] For those that just want to find out more about what AI is and the nature of the boom, I would recommend this discussion between Paul Krugman and Paul Kedrosky. 

[2] There would be a risk if a stock market bust hurt institutional investors. However, as Adam Tooze notes here: "the current bubble is a self-feeding surge in confidence, amplified by endogenous credit creation with retail investors and the media to the fore. The risks to the financial system are somewhat moderated by the fact that institutional investors are taking more cautious positions. If there is a bust, this is good news, at least so long as retail investors are not as heavily leveraged as professional hedge funds might be."