Oaktree Capital co-founder Howard Marks, who warned investors about dot-com excess before the 2000 crash, told Bloomberg in December 2025 that U.S. stocks are showing signs of the “early days” of a bubble driven by artificial intelligence. Three separate academic studies published in 2026 now back that warning with econometric evidence, applying statistical tests for explosive price behavior to AI-linked equities and finding patterns that mirror the speculative peak of December 1999 through March 2000. For the tens of millions of Americans whose 401(k) and target-date retirement funds are heavily weighted toward the same AI-heavy names, the convergence of a veteran market caller’s instincts and fresh quantitative research raises a pointed question: how much of their nest egg is riding on a repeat of history?
Marks’ dot-com track record meets 2026 AI exuberance data
Marks built his reputation by flagging speculative excess early enough for investors to act. In a Bloomberg interview shortly after his December 2025 memo, he drew a direct line between the current AI rally and the late-1990s tech mania, describing what he saw as matching psychological dynamics: widespread certainty that a new technology would change everything, coupled with a willingness to pay any price for exposure.
His comments gain sharper edges when read alongside a pair of 2026 arXiv preprints that test for bubble conditions using formal econometric methods. One study, available on the arXiv server, builds a model designed to separate legitimate adoption of a general-purpose technology from speculative price dynamics. That paper explicitly identifies the speculative peak of the dot-com era as running from December 1999 through March 2000 and uses it as the benchmark against which current AI equity behavior is measured. A second preprint, posted as AI bubble research, proposes a framework for date-stamping periods of exuberance under time-varying volatility and applies it directly to AI-linked stocks.
A third paper, published as of May 2026 and titled Boom, Bubble, or Buildout?, takes a multi-method approach to diagnosing whether AI is in an ongoing financial bubble. That study frames the central tension as a contest between two hypotheses: that the rally reflects durable technology adoption, or that it has crossed into speculation. Together, the three papers supply the kind of structured evidence that Marks’ qualitative warning lacked on its own. Instead of relying solely on analogies to the late 1990s, they attempt to quantify when price paths diverge from fundamentals in a way that historically has not been sustainable.
How AI concentration channels bubble risk into retirement accounts
The reason this research matters beyond trading desks is structural. Major U.S. index funds and target-date retirement vehicles allocate heavily to the largest technology companies, many of which have been reclassified by the market as AI plays. When econometric models detect explosive price dynamics in those same names, the risk does not stay confined to hedge funds or day traders. It flows directly into the default investment options that most American workers use for retirement savings.
The hypothesis that follows from the 2026 research is straightforward: if AI equities that exceed the exuberance thresholds identified in these studies are also the dominant weights in broad market indices, then a bubble in AI does not look like a sector-specific event. It looks like a market-wide vulnerability embedded in 401(k) plans, IRAs, and college savings accounts. Even investors who consider themselves diversified because they own “the whole market” through index funds may, in practice, be highly exposed to a narrow cluster of AI-driven valuations.
Target-date funds, which automatically shift from stocks toward bonds as savers approach retirement, do mitigate some risk by reducing overall equity exposure over time. But within the equity sleeve, they typically track the same capitalization-weighted benchmarks as standard index funds. That means that when a handful of AI-linked giants command a rising share of total market value, older workers nearing retirement can still find an outsized portion of their remaining stock allocation tied to the fate of a single theme.
For plan sponsors and individual savers, the policy implication is not necessarily to abandon AI-related equities, which may well reflect genuine productivity gains. Instead, the emerging evidence suggests a need to examine how much of a portfolio’s risk budget is unintentionally concentrated in one story. Simple steps such as capping single-stock weights, adding strategies that equal-weight sectors, or increasing exposure to assets whose returns are less correlated with AI can all reduce the impact if exuberance gives way to a prolonged drawdown.
Marks’ warning and the new econometric work do not guarantee an imminent crash. What they do is narrow the range of plausible narratives. When both seasoned judgment and formal models point to price behavior that rhymes with past bubbles, the burden of proof shifts. Investors and fiduciaries must be able to articulate why “this time is different” in more than slogans-and adjust retirement portfolios so that, if it is not, a generation’s savings are not learning that lesson the hard way.
Free tool for readers: Not sure whether your own retirement is on track? You can check your free Retirement Safety Score — a 0–100 number plus a few personalized steps — in about five minutes, with no sign-up required to see your score.
David M. Keller is a finance writer based in Columbus, Ohio, covering personal finance and consumer-focused economic topics. He earned his degree in journalism from Ohio University and began his career reporting on local business and economic trends for a regional media outlet. Since then, he has contributed to a variety of online publications, focusing on clear, practical coverage of topics such as cost of living, debt, and everyday financial decision-making.
