The title of this Insight might be the wrong question to ask. Investment bubbles have formed countless times in the history of financial markets. Their creation is not the primary cause of concern—the manner in which they are resolved, however, is. Thus, the more pertinent question facing investors is whether the artificial intelligence (AI) sector of the stock market is a bubble in the process of popping.

Indeed, it is when a bubble pops that the damage is done, both to the economy through decreased capital spending, rapid unemployment increases, and decline in overall consumption, and to the financial markets as capital losses lead to extended periods of reduced risk-taking and moribund capital creation. Neither of those outcomes is apparent at present. In short, the current environment is not what a bubble popping looks like. Let us add a little reasoning to that conclusion.

How does AI compare to prior bubbles?

First, a little historical context. Large language models at the core of the AI trade have been on investors’ radar for only three years. DeepMind at Google, Watson at IBM, and various university laboratories have been at work for decades, but it wasn’t until the introduction of ChatGPT by OpenAI and partner Microsoft in late 2022 that AI’s true capabilities became apparent. As we close 2025, the intervening period has seen the creation of multiple competing AI models, extensive infrastructure construction (i.e., data centers filled with AI-focused semiconductors), and tremendous capital deployment by legacy technology companies, private capital, and sovereign nations. In turn, returns on publicly traded AI-affiliated stocks have risen meaningfully since the 2022 bear market, leading many to wonder whether they have soared too high too quickly.

At Cerity Partners, we believe that the last three years are not long enough for a bubble of sufficient size and internal pressure to have formed. One thing to consider is how long it took previous revolutionary technologies to reach oversaturation. Steam locomotives were created around the turn of the 19th century. Yet it wasn’t until 1873 that a railroad construction overbuilding cycle led to a U.S. stock market crash, along with bankruptcies for many railway companies and the banks that financed them. Similarly, radio, electrification, and automobiles were all 19th-century inventions that did not achieve mass-market commercialization until the 1920s. Investor enthusiasm for these new technologies had a role in the stock market inflation of that decade. These industries were not the primary cause of the 1929 crash, though, which was driven by easily accessible margin debt that fueled a speculative bubble in the general stock market. It is reasonable to expect that in the modern digital age, industries evolve at a faster pace. Still, three years seems much too short to suggest the end is nigh for AI. It is noteworthy that AI capabilities are regarded as a strategic imperative by the world’s largest nations, as evidenced by their centrality to China–U.S. trade tensions and intense campaigning by Middle Eastern nations to access cutting-edge AI chips. If AI were a bubble, that would likely not be the case.

Furthermore, while the infrastructure is in a rapid construction phase, the end-use penetration of AI is just getting started. Given how often it is mentioned in earnings calls and the media, it seems likely the bigger risk is that companies do not develop the talent to use this new technology and are leapfrogged by competitors who do. While much of corporate America is still determining AI use cases, the idea that it will suddenly and unanimously abandon the technology defies logic, given AI’s significant impact on productivity and profitability.

What about debt financing and other factors?

Beyond the rationality of AI adoption, there are questions about the costs of building the data centers and buying the semiconductor chips that are so vital to the technology. These questions seem to fall into two categories: the use of debt financing and the circular nature of investments made by the leading companies in one another.

For the first of these, debt financing, there are many aspects to consider. To start, investors in technology companies aren’t used to debt capital. Meta, for example, had no debt on its balance sheet as recently as 2018. In the past three years, its outstanding debt has more than tripled to $49 billion.1 Investors have had to shift their mindset regarding technology companies from “asset-light” entities to ones with a sizable amount of physical property. That transition takes time. Also, whereas these companies have enjoyed very large and rapidly increasing cash flows, now those cash flows are being consumed by the capital expenditures they are making on these hard assets. The flip from positive to negative free cash flow has unnerved investors. But that is exactly what these companies should be doing: finding and investing in profitable business ventures. Were they not deploying their internally generated capital in this productive fashion, criticism might likely focus on their lack of imagination in finding new creative technologies. What’s more, the cost of the bonds they are issuing is relatively low, at 5% to 6% issued yields to maturity for companies like Oracle that have been in the crosshairs of concern recently.2 Finally, while the media is prolific in writing about the almost $150 billion in new debt issued this year by public companies pursuing the AI build-out,3 this total is paltry in comparison to the market capitalization of the major hyperscalers, the biggest five of which—Alphabet, Amazon, Meta, Microsoft, and Oracle—have a combined public equity valuation of over $10 trillion.4

A similar analysis applies to the so-called circular nature of AI investments. An often-cited example is Nvidia’s investment in AI pioneer OpenAI, which in turn has contracted with Oracle to host the cloud infrastructure powering OpenAI’s ChatGPT. Oracle then buys AI chips from Nvidia, completing the capital circuit. Two things stand out in this arrangement: At present, each transaction creates a positive return on investment. This differs from the rapid fiber-optic build-out in the telecom industry during the late 1990s, the result of which was that 75% of the laid fiber was “dark” or not generating revenue.5 Also, if one focuses on Nvidia as an important point in the circle, its investments into the AI ecosystem are a wise use of its capital. The company will likely generate over $110 billion in profit in its current fiscal year. That is a princely sum, one that raises the question of what the company should do with it. They could buy back shares, issue a dividend, or invest back into the very industry from which they derive such benefit. It is far more productive and profitable for Nvidia to invest. And if, as projected, OpenAI goes public at the highest valuation ever for an initial public offering, Nvidia’s return on the investment is likely to be quite impressive.

Another question that has arisen is whether AI companies are properly accounting for the speed at which the hardware they buy may become obsolete. Cloud providers are extending the depreciable life of graphics processing units (GPUs), which are AI-specific chips. Doing so lifts near-term earnings by spreading the cost across additional years, but anecdotes from Google and others suggest this is an accurate accounting of the GPU’s useful life, driven by aligning older chips to more mundane tasks and amplifying their usefulness through software improvements. While some may worry this resembles accounting maneuvering, the more meaningful risk may be that longer useful lives could slow the upgrade cycle for accelerators, tempering pricing power and the steep growth trajectory of leading chipmakers. This is a risk worth paying attention to. However, it is unlikely that this is as big a risk as media-friendly short sellers have recently promoted. We are paying close attention to this issue.

One last concept that is important to the discussion of whether we are presently seeing an AI bubble pop: Equity valuations are not a convincing argument that a major decline is due. Yes, at forward earnings multiples approaching 30x, they appear expensive.6 Yet relative to an annual earnings growth rate running from the high single digits to high teens in percentage terms, they may not be overpriced.7 If one compares these multiples to those reached during the dot-com bubble of the late 1990s, one might take comfort that the bellwether companies of that era often traded at multiples above 100x. That’s what a bubble looks like before it pops. None of this is to say that share price returns won’t be volatile in the sector.

To summarize, we don’t think the stock market action in the fourth quarter of 2025 shows evidence that AI is a bubble popping. We are increasingly confident that AI represents a generational technological shift with broad and lasting economic impact. But confidence in the long-term destination does not eliminate risk in the short-term path. Rapid innovation, changing depreciation dynamics, intense capital inflows, and uneven adoption mean mispricing will be common and volatility will be elevated. The prudent approach is clear: Maintain measured optimism, stay diversified, scrutinize cash-flow durability, and remain disciplined as markets oscillate between enthusiasm and reality. We believe that global stock markets are likely to expand over the next 12 months on the back of growing economies, rising profits, lower short-term interest rates, and greater certainty around trade policy. For these reasons, we think that not only AI-related technology stocks but also the broader array of stock sectors and industries can give positive returns.

If you have any questions or concerns, reach out to your Cerity Partners advisor or request an introduction today.

  1. Meta SEC 10-Q fillings September 30, 2025, and June 30, 2022. Debt is the total of long-term bonds and operating leases. ↩︎
  2. Oracle Corp. Final Pricing Term Sheet dated September 24, 2025. ↩︎
  3. Bank of America Investment Research dated November 17, 2025. ↩︎
  4. “The Wall Street Journal” market quotes as of November 19, 2025. ↩︎
  5. “The Wall Street Journal,” September 26, 2002. ↩︎
  6. Earnings multiples and growth rates on Alphabet, Amazon, Google, Meta, and Oracle as of November 19, 2025, per FactSet. ↩︎
  7. Ibid. ↩︎

Please read important disclosures here.