Warren Buffett’s cautionary words from 1999 are finding a fresh audience as artificial intelligence valuations push equity benchmarks toward levels not seen since the dot-com era.
The lesson Buffett delivered during a series of talks at the height of the internet boom, later distilled in a Fortune Magazine article, drew a distinction between a technology’s social reach and its capacity to reward shareholders. According to Buffett, the decisive question for investors is not how profoundly an industry will reshape society or how fast it will expand, but whether any single company holds a competitive advantage and, critically, how long that advantage can last.
He illustrated the point with the American auto industry, which at one stage counted roughly 2,000 manufacturers. Anyone who foresaw the automobile’s rise, he argued, would have assumed they had found the road to riches, yet by the 1990s only three automakers remained and none had proved a particularly rewarding investment. Aviation followed a similar arc: it collapsed global travel times from days to hours, but Buffett once described airlines as a “bottomless pit” for capital, noting that cumulative industry profits through 1992 amounted to zero.
The parallel to today’s AI buildout is not lost on market watchers. Data-center construction is capital-intensive, and with compute providers and frontier labs locked in fierce competition, it remains unclear where durable profits will ultimately settle. The winners, as with earlier technological revolutions, may only be identifiable in hindsight.
Buffett’s framework nonetheless offers a test investors can apply. Berkshire Hathaway (BRKA) (BRKB) took an initial position in Alphabet (GOOG) (GOOGL) in 2025, and Greg Abel has since expanded it into one of the conglomerate’s largest equity holdings with further purchases in 2026. Alphabet’s AI franchise appears to carry a moat that has supported aggressive spending on compute capacity, spanning chip design, cloud infrastructure, the Gemini model family, consumer applications including Search and the Gemini app, and its advertising back end.
Google Cloud CEO Thomas Kurian has said the payback period for its servers was under two years, with custom Tensor Processing Unit servers recouping the investment in half that time, a pace well ahead of most rivals. By contrast, neocloud operators selling commodity-like servers lack such protection, and the prices they command are likely to fall as compute supply constraints ease. On the software side, Anthropic and OpenAI lead on model quality today, but holding that position demands heavy continuing outlays while they compete against far better-capitalized companies with highly profitable businesses outside AI.
The stakes are considerable. Companies tied to artificial intelligence now represent roughly half of the S&P 500’s market capitalization, and trillions of dollars in forthcoming bond issuance is helping drive the U.S. economy.
At Berkshire’s 2026 annual meeting, Buffett told CNBC that markets resemble a church with a casino attached, with the church standing for fundamentals-driven long-term investing and the casino for short-term trading. The casino, he said, has become very attractive to people, adding: “[T]hat’s not investing, it’s not speculating, it’s gambling.” He tempered the remark by noting that this does not make investing terrible, but it does mean “prices for an awful lot of things will look very silly.”
Valuation gauges support the concern. The S&P 500 has posted more than two dozen record highs in 2026, and the cyclically adjusted price-to-earnings ratio, which smooths earnings over ten inflation-adjusted years, has climbed above 40 for only the second time in market history, the first being the dot-com peak. The measure also spiked ahead of the Great Depression in 1929 and peaked in late 2021, months before the index entered a bear market. A reading this elevated does not signal an imminent crash, but it does indicate unusually rich pricing.
IndicatorReadingContextS&P 500 Shiller CAPE RatioAbove 40Second time on record; first was the dot-com bubbleS&P 500 record highs in 2026More than two dozenReflects sustained rally through the yearAI-linked share of S&P 500 market capAbout 50%Concentration risk for index investorsU.S. auto manufacturers at peakAbout 2,000Only three remained by the 1990sCumulative airline industry profit through 1992$0Buffett’s “bottomless pit” example
Note: CAPE readings and index milestones reflect conditions as of mid-November 2026.
History offers a counterweight to the gloom. Since 2000 the S&P 500 has absorbed the dot-com bust, the Great Recession, the COVID-19 crash and the 2022 bear market, along with numerous smaller corrections, and is up nearly 750% over that span. Expensive markets still contain reasonably priced and even undervalued businesses, and quality companies have reliably compounded value for patient holders.
Near-term direction is impossible to call. What the 27-year-old warning supplies is a filter: in an era when AI touches nearly every sector, the question worth asking is not which technology will change the world, but which company can defend its edge long enough to be paid for it.