For the first time since the dot-com crash, fewer students are studying computer science. Across the University of California system, 12,652 undergraduates are majoring in CS — down 6% from last year, down 9% over two years. A national survey by the Computing Research Association found 62% of programs reporting similar declines.
Students aren't leaving technology. They're leaving the discipline that built it.
Where They're Going
UC San Diego is the only campus in the system where CS enrollment grew. The difference: UCSD offers California's first undergraduate AI major. One in five applications to the department's CS programs now targets that AI track. Professor Steven Swanson called it "a great start" and expects the fraction to grow.
MIT's AI and Decision Making major, launched in 2022, is now the institute's second-largest program with nearly 330 students. SUNY Buffalo's AI master's program grew from 5 students to 103 between 2020 and 2024 — a 20x increase. Nationwide, there are now 193 bachelor's and 310 master's programs specifically in artificial intelligence. Most didn't exist three years ago.
The University of South Florida enrolled 3,000 students in its new AI and cybersecurity college in fall 2025. UC Santa Barbara's admissions director, Cuka Acosta, summarized the shift: "The biggest surprise is us trying to fill computer science."
Why They're Leaving
The math is straightforward. Computer science graduates face 6.1% unemployment — higher than biology or art history majors. Underemployment sits at 16.5%. Entry-level tech roles have dropped more than 50% from pre-pandemic levels. Junior developer job postings fell 73% between 2023 and 2025. CS degrees nearly doubled in a decade, from 51,696 graduates in 2013-14 to 112,720 in 2022-23, flooding a market that simultaneously contracted.
Students read the same headlines everyone else reads. Spotify's senior developers haven't written code in two months. Google's Gemini can pass reasoning benchmarks that stumped PhD researchers. Companies that laid off 20% of their engineering staff are telling investors that AI writes most of their new code.
The rational response, if you're 19 and choosing a major, is to study the technology doing the replacing rather than the discipline being replaced.
The Irony
There's a problem with this logic. AI programs teach students to use machine learning frameworks, prompt large language models, and apply statistical methods to datasets. They generally do not teach compiler design, operating systems, network protocols, or memory management — the substrate on which every AI model runs.
UCLA professor Glenn Reinman put it directly: "Generative AI has tremendous potential to transform industries, including computer science, but it does not replace the need for trained humans."
The people building the next generation of AI infrastructure — the actual training frameworks, the chip architectures, the distributed systems — need deep computer science knowledge. The people building applications on top of AI need much less of it. Students are betting that the application layer is where the jobs will be. They might be right. But someone still has to build the layer underneath.
The Cycle
This has happened before. In 2002, CS enrollment crashed after the dot-com bust. It took nearly a decade to recover. By 2015, enrollment was growing so fast that universities couldn't hire faculty quickly enough. The boom produced a glut of graduates that peaked right as AI tools started automating the entry-level work those graduates were trained for.
Now the cycle is compressing. The bust in CS enrollment and the boom in AI enrollment are happening simultaneously, driven by the same cause. Students aren't responding to a recession — they're responding to a perceived permanent restructuring of what software development means.
UCSD professor Leo Porter identified the core challenge: "If we don't shift our curriculum, then they probably won't" adapt to a world where generative AI is a standard tool.
What This Actually Means
The pipeline that produced the software industry's workforce for two decades is contracting. Not because technology is declining, but because students believe the specific skill set a CS degree teaches — writing code by hand, understanding systems from first principles — is less valuable than knowing how to direct AI systems that do it for them.
If they're right, the software industry's next generation will be operators, not builders. They'll be exceptionally good at using tools they don't understand. That works fine until the tools break, or until someone needs to build the next tool, and the people who know how have all retired or switched careers.
The Bureau of Labor Statistics still projects 22% growth in computing occupations through 2033. But "computing" increasingly means something different than what a CS degree teaches. The question isn't whether the jobs will exist. It's whether the people filling them will know how computers actually work — or whether that will matter.\n\n---\n\n*If you work with AI tools daily, check out my AI prompt engineering packs — battle-tested prompts for developers, writers, and builders.*
Top comments (0)