Radio ExpressTV
Live
The AI Bubble Will Not Ignite a Financial Crisis
A familiar sense of trepidation has returned to financial markets. Amazon is committing $100 billion to data centers, while Meta has pledged over $600 billion to build data centers over three years. Microsoft, Google, and Apple are also planning to spend additional hundreds of billions, with AI investments soaring towards trillions. Are we witnessing the emergence of a bubble, and what might happen if it bursts?
We’ve been through this before. One of us (Martin) served as Chairman of the Council of Economic Advisers in the White House during the tech boom of the late 1990s, and the other has observed California deal with the collapse of internet companies and the 2008 financial crisis from various governmental and advisory positions. Our experiences suggest that even if a correction occurs in the AI sector, it won’t trigger a financial crisis like the one that devastated the economy in 2008-2009.
The reason is that the investment structure in AI is fundamentally different from what we saw during those previous events. The internet companies’ crash is remembered for its massive losses—the Nasdaq index plummeted by 77% from its peak in March 2000—and remarkable shutdowns like Pets.com. However, the economic story was much more complex.
Yes, there was excessive investment in internet infrastructure, and many big bets turned out to be premature. But this “overinvestment” left behind an infrastructure of exceptional value. The fiber-optic cables laid during the boom enabled the emergence of a broadband economy. The server capacity that seemed excessive in 2001 became essential by 2005. Productivity growth remained surprisingly strong even after the bubble burst, demonstrating that the underlying technology was transformative, even if the initial business models were not.
It’s crucial to note that the internet companies’ collapse did not threaten the financial system overall. Speculation was primarily funded through stock markets, not debt. When valuations collapsed, shareholders lost money, but banks remained able to meet their financial obligations. There was no wave of defaults, no credit freeze, and no need for massive government bailouts.
In contrast, the 2008 crisis was about how assets were financed. High-risk mortgages were bundled into securities, sold to pension funds and international investors, and used as collateral for further borrowing. The entire global financial system became a house of cards built on rising home prices.
When those prices fell, the web of complex obligations collapsed: banks found they owned worthless securities, credit markets froze, and what began as falling home prices morphed into a global financial crisis that required unprecedented governmental intervention. The problem was not just overvaluation but a mix of leverage, complexity, opacity, and the spread of housing risks throughout the financial system.
The AI boom is different. The trillions being spent on core AI infrastructure primarily come from the balance sheets of the most profitable and wealthiest companies in the world. Companies like Apple, Microsoft, Google, Amazon, and Meta do not borrow money to build their AI capabilities. When Apple commits to spending $500 billion over four years on core AI infrastructure projects, it is not issuing debt securities that must be repaid regardless of revenues. This is capital-funded investment by one of the highest-valued companies in history. If earnings disappoint, Apple’s share price may take a hit, but the company will not default, nor will it trigger a cascade of failures in the financial system.
Similarly, the specialized nature of AI infrastructure limits contagion risks. Data centers are not converted into securities and sold to pension funds. AI chips are not used as collateral for financial derivatives. Of course, a sharp correction in tech stock valuations may occur, and companies may cut back on AI investments if returns prove disappointing. But does any of this lead to a financial crisis akin to 2008? Structural differences suggest otherwise.
This does not mean that AI poses no economic risks. It means that these risks do not primarily threaten financial stability. Concerns worthy of attention include: failure to achieve significant productivity gains leading to slower growth and disappointment; rapid displacement of workers requiring robust policies to support education and training; economic power concentration raising questions about antitrust; and pressures on electrical grids and construction capacity that could limit growth and increase household electricity bills.
Even if AI valuations are frothy and current revenue expectations overly optimistic, the infrastructure being built holds lasting value. These data centers will not disappear. The fiber-optic connections being laid are permanent additions to our digital infrastructure, and trained AI engineers represent human capital that will continue to generate value for decades to come.
If someone had told you in 1999 that America would “lavishly spend” hundreds of billions of dollars on internet infrastructure, you might have worried about a bubble. But that overinvestment enabled the entire digital economy—YouTube, Netflix, cloud computing, and remote work. The societal return was substantial, despite many companies failing and investors losing money.
Our understanding that speculation on AI is unlikely to lead to a financial crisis means policymakers must prioritize correctly. Instead of overregulating to prevent a crisis that is not forthcoming, they should focus on risks that threaten the real economy: worker displacement, algorithmic bias, data privacy, and competition dynamics. They should encourage investments in complementary infrastructure such as electrical power, cooling systems, and trained workers. They should prepare for labor market adjustments by enhancing unemployment insurance, retraining programs, and transferable benefits.
Moreover, they should monitor rising debt levels, as the only risk that might alter our analysis is if AI assets become heavily debt-backed or converted into securities. Financial regulatory bodies should accordingly monitor banks and pension funds.
Transformative innovations typically take decades to fully spread across economic sectors. It took 40 years for electricity to revolutionize manufacturing; 20 years for the internet to reshape retail. If AI proves similarly transformative, the current speculative investments may turn out to be prescient.
The 2008 financial crisis taught policymakers to be vigilant about financial stability risks. But the differences between asset classes and funding structures are significant. The AI boom may be speculative. It may not be without exaggeration. It may even be a bubble. But it does not pose a threat to the financial system.
