What connects $400 billion of gleaming new data centers, thousands of newly minted AI unicorns, and a rising wave of tech layoffs? To some industry veterans and economists, they represent the symptoms of a bubble swelling at breakneck velocity a bubble that may shortly deflate with a crash.
The comparisons to the late-1990s dot-com bubble are hard to escape. At the time, firms were able to prefix “.com” to their names and see valuations skyrocket. Now, the simple mention of “AI-powered” can fling open the wallets of investors. There are now over 1,200 unicorns around the world, according to Founders Forum, and 65 new U.S.-based AI unicorns were created in 2024 alone most worth billions even though they are less than five years old.
The capital spending figures are astronomical. Microsoft and Amazon each will spend roughly $100 billion on AI next fiscal year. Alphabet has set a goal of $85 billion, and Meta is estimating anywhere from $66 billion to $72 billion. Combined, the four behemoths are set to surpass $400 billion in AI-related capex more than the European Union spends on defense in a quarter. These resources are pouring into densely packed data centers full of Nvidia GPUs, sophisticated networking equipment, and water and electricity-guzzling cooling systems. As Microsoft CFO Amy Hood said, “We will continue to invest against the expansive opportunity ahead.”
But some experts caution that the “opportunity” could be more hype than substance. Apollo Global Management’s chief economist Torsten Slok has also pointed out that the current S&P 500’s top 10 firms are more overbought compared to the height of the 1990s IT bubble. Adding to the danger, more than half of world internet traffic is now created by automated bots with 20% being “bad bots” that can both pump up engagement metrics as well as advertising revenues. Inflated metrics can deceive investors regarding actual user adoption, similar to the vanity metrics that drove previous bubbles.
The economic impact of AI investment is already evident in macro statistics. Renaissance Macro Research analyst Neil Dutta estimates that IT spending on AI has added to U.S. GDP growth over the last two quarters more than all consumer spending put together. Others label it a “private-sector stimulus program” supporting the economy in the face of tariff headwinds. But that begs an uncomfortable question: what if the cycle of AI investment were to slow down or even turn around?
Labor market signals add another layer of complexity. Challenger, Gray & Christmas reports that for the first seven months of 2025, generative AI adoption accounted for more than 10,000 job cuts, placing AI among the top five causes of layoffs this year. The tech sector alone has announced over 89,000 cuts, up 36% from 2024. Since 2023, more than 27,000 job losses have been directly tied to AI deployment. Early effects are focused on jobs such as customer service, marketing, and administrative support high-exposure jobs to automation.
Economic modeling indicates that widespread AI adoption may increase developed market labor productivity by approximately 15%, but increase unemployment by about 0.5 percentage points in transition. Goldman Sachs Research estimates that 6-7% of U.S. jobs may be displaced under base case assumptions, with greater threats facing programmers, accountants, and legal assistants. Historically, displacement impacts decline within two years, but a steep, front-loaded adoption curve may increase short-term dislocation.
For the moment, high-level labor statistics indicate no meltdown. Multivariate analyses with various measures of AI exposure conclude that unemployment has increased for highly exposed and less exposed occupations alike, but more so for the latter. The cooling is more pronounced in the patterns of hiring, however: initial corporate job ads have declined 15% year-on-year, and the proportion of employers listing “AI” among job requirements has risen 400% over two years.
The longevity of the AI boom will depend on the technology producing commensurate top-line growth. To date, the largest beneficiaries are infrastructure providers cloud providers and chipmakers more than the AI applications themselves. Most consumer-oriented AI solutions, particularly chatbots, are still expensive to maintain and have yet to demonstrate sustainable profitability. If investor excitement stalls prior to monetization getting ahead of it, the correction may match or surpass the dot-com bust in magnitude.
As Geoffrey Hinton, Nobel laureate and AI pioneer, remarked in a different context, “We’ve never had to deal with things more intelligent than ourselves before.” In the financial realm, the challenge may be simpler but no less urgent: ensuring that the intelligence driving capital allocation is not blinded by its own exuberance.
