Stanford economist Jacob Light links computer science enrollment reversal to AI era

Computer science enrollment fell during the 2025-26 academic year for the first time in roughly two decades, according to new research by Jacob Light, an economist at Stanford University's Hoover Institution. The study does not estimate how big the drop would be in percentage terms nationwide. But it does identify something just as important for decision-makers: the long-running boom in CS enrollment share appears to have flattened and reversed.

Light’s analysis uses a dataset covering more than 50 million course sections offered since 1996 across 1,019 US colleges and universities. Within that sample, he reports that at the average school, enrollment in computer science classes fell about 4.6% year over year, and computer science’s share of overall course enrollment dropped 6.3 percentage points. Light describes the “most notable recent development” as “the flattening and reversal in the Computer Science enrollment share trend,” and he says that in 2025-26 CS enrollment share declined for the first time in roughly two decades, interrupting a period of sustained and unusually rapid growth.

So is this the “ChatGPT caused it” moment? Not exactly. Light explicitly cautions that the timing overlaps with AI adoption, but the data cannot establish causality. In the study, Light treats enrollment patterns as descriptive evidence rather than an estimate of the causal effect of ChatGPT or AI exposure on student demand. That distinction matters because executives, boards, and investors will see a headline like “CS enrollment fell” and immediately try to connect it to one storyline. Light is basically telling you to slow down: the reversal is real in the data, but the driver could be multiple things happening at once.

The second proof point is coming from national enrollment reporting. Separate data from the National Student Clearinghouse Research Center shows undergraduate enrollment in Computer and Information Sciences and Support Services at four-year colleges fell 8.1% year over year in the fall of 2025. It dropped from about 659,700 students in 2024 to 606,100 in 2025. Even with that decline, enrollment remained well above 2022 levels of 574,333 students, which suggests universities are not watching a collapse from the bottom of the trend. Instead, it looks like a cooldown after years of rapid growth.

Why would that matter to anyone outside a faculty meeting? Because computer science has been one of the most reliably “hot” major categories in the US. Light’s research notes that computer science’s popularity surged as demand for software engineers and other tech workers increased. That boom is visible in degrees too. The number of bachelor’s degrees awarded in computer and information sciences more than doubled over the past decade, rising from about 56,000 in 2014 to about 122,000 in 2024, according to data from the National Science Foundation’s National Center for Science and Engineering Statistics. Now, the first dip in roughly two decades raises uncomfortable questions for university planners and for companies that depend on CS pipelines.

Light lays out five leading explanations for the enrollment drop, and this is where the AI debate meets real-world incentives. The possibilities include a weaker entry-level software engineering job market, uncertainty about the value of computer science degrees in the AI era, weaker academic preparation among incoming students, changes in college demographics linked to immigration policy, and the expansion of competing fields such as data science. The key point is that Light’s dataset cannot distinguish among these possibilities. His approach is to observe the pattern, quantify the shift, and then wait for additional cohorts to make their major and course choices in the AI moment.

For executives and boards, the practical takeaway is not “ChatGPT killed CS.” The practical takeaway is that the enrollment system is not immune to shocks in perception, labor-market expectations, and curriculum competition. Even Light says computer science has historically been a boom-and-bust major, and a decline would not be new for universities. But he also signals that there’s reason to believe CS enrollment may decline in the short term, even if many graduates may still find value in a CS degree. The more important line is his forecast of clarity: it will become clearer “as additional cohorts make course and major choices.”

Meanwhile, the broader ecosystem is already moving. Generative AI assistants can write code, debug programs, and complete technical assignments, and that has fueled debate about whether the traditional value proposition of computer science changes in an AI era. If students begin to expect AI to do more of the grunt work, their major choice could shift toward other pathways, including data science or other skills-adjacent tracks. And if the entry-level job market is perceived as less welcoming, even strong demand narratives may lose their pull.

The strategic stake is simple: who wins the talent pipeline when the pipeline’s baseline starts to dip? Universities will need to interpret whether 2025-26 is a temporary dip or a new regime, especially because the CS enrollment share reversal interrupted a sustained period of unusually rapid growth. Companies that rely on new grads should treat this as an early warning that perceptions of “what CS is for” are under negotiation. For decision-makers, the next phase is about monitoring cohorts, tightening the connection between skills taught and roles offered, and being honest about what AI changes and what it does not. Light’s research gives you the signal. The cause is still up for debate. But the direction of the trend is no longer hypothetical.