Mises Wire

The Current AI “Investments” Are Turning into a Bubble

AI bubble

There’s no question that we are presently living through a time of major technological upheaval and advancement. I’m somewhere between the extremes when it comes to long-term views—I’m not convinced that AI will kill us all, nor am I convinced that AI will usher in some new post-scarcity utopia.

I’m more confident about the shorter run: we are in a big AI bubble.

There has been a huge boom in the industry in the last few years. The question is whether that boom is sustainable or whether it will turn to bust.

The Artificial Boom in Artificial Intelligence

Consider a recent headline: on November 10th, the Wall Street Journal reported that “OpenAI Continues to Suffer Heavy Losses as Revenue Rises to $10 Billion.” Despite some revenue growth “from the second to the third quarter. . .the company’s adjusted operating losses ticked up slightly from $10.1 billion to $10.9 billion.” The article compares OpenAI and Anthropic’s financial positions. Anthropic is doing better than OpenAI, but only by suffering smaller losses.

The two companies are competitors in a larger field, which requires tough decisions like cutting prices and making big investments to provide more and better products. The WSJ reports that OpenAI recently reduced the prices of its newest models and that while the number of paying users has increased, non-paying users outnumber them over 10-to-1. It seems the massive investments are struggling to find enough paying customers to justify themselves.

The companies compete for paying users and for financing. Due to recent investor skittishness, OpenAI “raised less than half of its initial $30 billion target” in its last funding attempt. For the same reasons, Anthropic has downgraded its target valuation in an upcoming IPO from $2 trillion to $1.33 trillion.

The competition extends to the factors of production as well. Google, Microsoft, OpenAI, Anthropic, xAI, Meta, Amazon, and Apple have been in a fierce Hungry Hungry Hippos battle over land, labor, and capital goods. Chief among the latter category are the GPUs and chips produced by Nvidia, AMD, and TSMC. Of course, all of the hardware needs to be powered, cooled, and put under a roof (those controversial data centers that everyone wants, maybe indirectly, but no one wants in their “backyard”).

Despite being made out of metals, silicon, and plastic, the hardware has a limited shelf life. Bloomberg and Reuters reported on November 6th that, “Microsoft is winding down an arrangement to rent AI computing from Oracle as part of a larger reassessment of billions of dollars in cloud partnerships.” Why? “The rented computing capacity became less valuable over time as it no longer met Microsoft’s evolving needs. . .the Oracle hardware used an older generation of Nvidia chips.” Competition requires new services, which require new hardware.

Coming back to financing issues, Reuters also reported on November 9th that, “Banks, investors grow wary of AI debt, spooked by market selloff,” a type of headline that is becoming increasingly frequent. Banks and investors have bought as much debt as they can stomach from big tech, but big tech debt issuance is projected to “reach $420 billion in 2027, more than 60% above its estimate of $250 billion for fiscal 2026.”

Putting this all together, we have fierce competition for buyers, factors of production, and financing. The buyers are already stretched thin—consumers and businesses have faced a prolonged period of price inflation. The hardware companies cannot churn out new hardware fast enough. And investors have forked over hundreds of billions of dollars while the big AI companies are asking for hundreds of billions more. Everybody is bumping into real resource constraints.

The underlying issue is that the current profits of the companies that supply the resources depend on the investment decisions of companies that sell AI products to end users, and these companies are suffering huge losses. TSMC profits when Nvidia and AMD buy their chips. Nvidia and AMD profit when hyperscalers like Google and Microsoft buy their hardware. The hyperscalers and data centers profit when AI companies like OpenAI and Anthropic rent their computing power and storage. The success of this entire structure depends on end user demand. So far it is unproven, and it seems unlikely that it will justify the trillions of dollars in investment.

AI and ABCT

Indeed, the Austrian theory of the business cycle is on full display. Easy credit has encouraged massive investment in AI, and only time will tell how much of it was malinvestment. Investment in AI has skyrocketed since 2020, only blinking a little when the Fed increased interest rates from zero in 2022-2023. By mid-2024, the Fed began decreasing interest rates again. These downs and ups in the fed funds rate mask what has happened with the money supply over the same time period, however. M2 has increased by more than $8 trillion since the end of 2019, an increase of over 50 percent. Much of that increase was channeled through credit markets, encouraging new, big, long-term investments in AI. For example, JPMorgan Chase has about $29 billion in outstanding CRE loans in the data center category—their exposure in this category recently surpassed the industrial and office building CRE categories.

Over the same time period, the AI-related companies mentioned in this article have seen huge increases in total market capitalization. The largest relative increases came from Nvidia (28.3x), AMD (15.7x), and TSMC (7.2x), which represent earlier-stage production. These figures are astounding (those are multipliers, not percent increases), especially since their growth is based on anticipated but unproven profitability later in the structure of production.

Rothbard points out that, “The capital-goods industries—especially the industries supplying raw materials, construction, and equipment to other industries—expand much further in the boom, and are hit far more severely in the depression.” Investors are already showing signs of doubt in Nvidia: its stock price fell 30 percent from October 6th to November 10th.

Will the bust be big or small? Sooner or later?

The Fed-enabled easy credit both before and after the 2020 crisis has created a bubble. The question is when and how it deflates. Will it be a big, calamitous pop? Or will it be a slower leak? Will it come sooner or later?

Answering these questions is beyond what anyone can confidently predict, but I can comment on some broader political and economic factors that could shape the bust. The Democrats just won majorities in the House and the Senate, and they might put up roadblocks to AI investments. The Fed continues to scramble through unpopular price inflation—they just recently signaled a tightening cycle with a small increase in the fed funds rate, the first increase since mid-2023. Household debt has increased by about $5 trillion since 2020, a 35 percent increase. Delinquencies on all types of debt have increased since 2022—credit card debt delinquencies are at 2009 levels and auto loan delinquencies are at the highest level ever. Will households choose groceries or the latest AI model when times get tough?

It’s difficult to imagine that the easy credit and yield-chasing investor attitude will continue to surge. When credit dries up, we should expect a re-evaluation of the malinvestments made during the boom, a “process by which the economy adjusts to the wastes and errors of the boom, and reestablishes efficient service of consumer desires. The adjustment process consists in rapid liquidation of the wasteful investments,” according to Rothbard.

I’m not saying that AI will die. Some firms will live through the bursting of the bubble, and the resources of failed firms will be sold to surviving ones, at liquidation prices. When the bubble bursts, the government should get out of the way and let it happen. It’s how scarce resources are put to the best possible use in the service of consumers.

image/svg+xml
Image Source: Adobe Stock
Note: The views expressed on Mises.org are not necessarily those of the Mises Institute.
What is the Mises Institute?

The Mises Institute is a non-profit organization that exists to promote teaching and research in the Austrian School of economics, individual freedom, honest history, and international peace, in the tradition of Ludwig von Mises and Murray N. Rothbard. 

Non-political, non-partisan, and non-PC, we advocate a radical shift in the intellectual climate, away from statism and toward a private property order. We believe that our foundational ideas are of permanent value, and oppose all efforts at compromise, sellout, and amalgamation of these ideas with fashionable political, cultural, and social doctrines inimical to their spirit.

Become a Member
Mises Institute