By: Adrien Matray
Research Economist and Associate Policy Advisor, Federal Reserve Bank of Atlanta
CEPR
Look at the numbers surrounding today’s AI investment wave, and it is hard not to think of the late 1990s. Tech giants are committing tens of billions to artificial intelligence. Startups in the space carry valuations that would have been unimaginable a few years back. Price-earnings ratios in the sector have climbed to levels we haven’t seen since the dot-com era.
The resemblance with the ICT boom goes deeper than stock prices. Then, as now, a genuinely transformative technology attracted massive capital inflows and triggered a fierce war for talent. Then, as now, the prevailing view was reassuring: even if things get frothy, the lasting benefits—faster internet, the rise of e-commerce, the digital infrastructure we all rely on today—will more than compensate for the excesses.
In a recent study with my coauthor Johan Hombert, I set out to test whether that optimistic story holds up when you look at what happened to the workers themselves. The answer, I believe, should unsettle anyone watching the current AI frenzy: many of the workers who flocked to the booming ICT sector ended up significantly worse off over the long run.
Using detailed French administrative data, we tracked thousands of high-skilled workers who entered the labor market in the late 1990s. Those who started their careers at ICT firms during the boom initially earned a wage premium—no surprise there. But fast-forward 15 years, and the picture reverses: these same workers were earning roughly 7% less than comparable peers who had begun in other industries. To put that in perspective, it is the equivalent of losing about two years of normal career progression.
A natural objection is that these workers simply chose the wrong firms—startups that went bust after 2001. But that’s not what we find. The wage penalty shows up even among workers who joined ICT companies that survived the downturn and continued to thrive. Nor is it a story about a sector in permanent decline: workers who entered ICT just a few years after the boom, once the dust had settled, experienced no such long-term penalty.
So what explains it? The key mechanism, we argue, is accelerated skill obsolescence. Technology booms are periods of intense experimentation. Workers acquire expertise in whatever tools and platforms are cutting-edge at the time, but the pace of change means those specific skills become outdated quickly. Think of the developers who mastered static HTML in the late ’90s, only to see their know-how lose its value once database-driven web applications took over. Or the IT consultants who built on-premise enterprise systems right before cloud computing reshaped the industry.
Crucially, we show that this effect was concentrated among STEM workers—engineers, developers, and technical specialists whose human capital was tightly tied to specific technological implementations. Workers in more portable roles like finance or general management at the very same firms did not suffer the same long-term wage decline.
There is another dimension to our findings that I find particularly striking. Capital during the boom did not flow randomly across ICT firms. It flowed disproportionately toward the companies engaged in the most aggressive experimentation—exactly the firms where technological turnover was fastest and skill obsolescence most severe. In other words, the financing boom didn’t merely expose more workers to this risk; it actively funneled them toward the positions where the risk was greatest.
What does all this mean for today’s AI boom? The parallels are, if anything, more acute. AI frameworks, model architectures, and deployment best practices are evolving not by the year but by the month. A large language model hailed as state-of-the-art today can feel outdated within a year. The half-life of specific AI skills may be even shorter than what we documented for ICT.
In our data, roughly one-third of skilled workers entering the French labor market during the ICT boom chose that sector. The share going into AI-related fields today may be smaller in relative terms, but the absolute numbers—and the capital at stake—are vastly larger. If the same dynamics are at play, we could be steering a generation of talented workers toward careers whose foundations erode faster than they realize.
None of this means the AI boom is a net negative for society. The technologies under development may well deliver transformative productivity gains and enormous value. But it would be a mistake to assume that what is good for aggregate innovation is automatically good for the individual workers powering it. When speculative financing meets rapid technological change, the labor market can guide talented people toward roles that come with impressive starting salaries but poor long-term prospects.
For policymakers, our work highlights that the labor market consequences of investment booms deserve far more scrutiny than they typically receive. For workers weighing a career in the hottest sector of the moment, it is worth asking whether those premium entry wages are compensating for genuine risk rather than simply reflecting high productivity. And for anyone following the AI revolution, our findings serve as a reminder that more investment and faster innovation do not automatically translate into broadly shared gains.
The dot-com era taught us that bubbles burst. What Johan Hombert and I have shown is that, even before they do, they can quietly erode the human capital of the very workers caught up in the excitement.
About the Author
Adrien Matray is a Research Economist and Associate Policy Advisor at the Federal Reserve Bank of Atlanta. His research focuses on empirical corporate finance, entrepreneurship, and innovation, with a particular interest in how financing constraints and technological change affect firms and workers. He is also a Research Affiliate at the Center for Economic and Policy Research (CEPR). For Adrien’s publications, visit https://adrienmatray.net/published-papers/
