The Recent Rise in Information-Sector Layoffs and What it Could Tell Us About AI

At The Budget Lab we examine incoming labor market data closely for a variety of purposes, including to identify any potential impacts of AI. One high-quality, timely source of such data is the Job Openings and Labor Turnover Survey (JOLTS), which releases April estimates this week. A recent trend from that survey caught our eye: the large and persistent rise in layoffs in the Information sector, shown in Figure 1. Given that a number of tech firms are represented in that industry—including Meta and Microsoft—could the layoff spike be a sign of AI disruption?

Figure 2 shows that hires are also trending up in the Information sector, from 2.7% in November 2025 to 3.4% in March 2026. Together, these trends could reveal an AI footprint in the sector—maybe employers are firing workers they don’t need in the AI era, while hiring the ones they do.   

When interpreting data like these, we do a few things. First, we look for reasons why the trend might not be “real”. For example, could it be due to sampling noise rather than economic signal? This seems unlikely given that the rise in layoffs started in December 2025 and exceeds the typical monthly variation, having been near or above its post-pandemic high since February 2026.1

Second, we look at whether other data tell the same story. Two natural points of comparison are the unemployment insurance claims data and the Current Population Survey unemployment rate, both of which come from entirely different sources than the JOLTS layoffs estimates.2 Interestingly, neither of these alternatives are fully consistent with the idea that there are AI-related layoffs in the Information sector. They are shown in Figure 3 on the left-hand side as the share of continued claims (roughly, the population of UI recipients) in the Information sector and in Figure 4 as the unemployment rate in that sector. If hiring is up along with firing, workers in that sector could be finding new jobs quickly and so not filing UI claims or staying long in unemployment. In other words, these data series could be consistent with individual firms firing workers who are being hired by other firms, but not with a decline in the labor needs of the entire sector.

Third, we want to understand whether the layoff trend might have reasonable explanations other than AI. If so, which mechanisms are the most likely given what we know? This requires getting more specific about what exactly is contained in the Information sector. It contains a wide variety of industries, from newspaper publishing and telecommunications to data processing and web search. What this means is that, out of the 66,000 Information-sector layoffs in March, many but not all were recorded by firms that would be recognizably “tech”.   

Even focusing just on layoffs by technology companies, it is not obvious that AI is the culprit. These layoffs could have other causes than AI, like higher borrowing costs or the overhang of previous overhiring. Distinguishing between potential causes will require different methods. For example, The Budget Lab tracks measures of occupational churn, both in the overall labor market and by sector. To date, Figure 5 shows that we don’t see any unusual increase in that churn for the Information sector, which one might expect to see if AI were causing labor market disruption. 

Another way to isolate AI effects, as opposed to other potential causes of labor market change, is to make statistically careful comparisons of AI-exposed and unexposed workers. This approach has its own significant downsides, but provides another way of grappling with the question. Here again, we do not yet find evidence of AI labor market effects. 

Our bottom line is frustratingly incomplete. It would be unwise to simply disregard this data point from the Information sector, which comes from a high-quality survey and is not obviously the result of some other factor—at least to our current knowledge. But it would be similarly unwise to leap to an unwarranted conclusion about AI impact, given the lack of corroboration in other datasets and the other potential explanations for the trend. 

The authors are grateful to David Ratner for insightful feedback on an earlier draft.