Article 347: US Computer Science Enrolment Fell 12 Per Cent This Autumn, and It Is the First Number in This Story That Is Not a Forecast
Early figures from the National Student Clearinghouse show undergraduate computer science enrolment at four-year US colleges down 12 per cent this autumn, with graduate numbers down 10 per cent. Total college enrolment barely moved.
That is not a rounding error and it is not a blip in one department. It is a specific, measurable retreat from one field while the rest of higher education held steady.
The obvious explanation is not the only one
The tidy story is that students read the same headlines you did - junior developer roles compressed, coding agents shipping, entry-level hiring thin - and made a rational choice about a four-year degree plus debt.
That is probably part of it. Three other things are also true and easier to verify:
- Cost and debt. CS is not unusually expensive, but it is often chosen for a high expected salary, and expected salary is exactly what becomes unstable when the entry-level rung gets crowded.
- Hiring pipelines tightened before the tooling arrived. Large tech hired heavily in 2021-22 and spent 2023-25 correcting. A student starting a degree in 2026 saw the correction, not the boom.
- The curriculum question. If the first job of a graduate is now "review the agent's work," then a degree built around writing the code from scratch is teaching something adjacent to the job.
Why the graduate number is the stranger one
A 10 per cent fall in graduate enrolment is harder to explain as a market signal. Master's programmes are usually counter-cyclical: when the job market is soft, people stay in school. A drop there suggests the calculation changed rather than the market merely wobbling - people deciding that one more year of CS tuition is not the thing that fixes their prospects.
What it means for the industry
Software is the industry that has spent two years arguing that AI will make engineers more productive rather than redundant. The labour market has not confirmed that yet. Neither has the educational pipeline, which is voting earlier and with less information than anyone. Universities feel a 12 per cent drop as budget, which means the response will be programme cuts rather than curriculum reform - the cheapest adjustment available.
Key Takeaway: every claim about AI and work in this newsletter has been, at best, a leading indicator from job postings or survey intentions. This is a register of what seventeen-year-olds actually did. They may be wrong - markets mis-price things constantly, and the AI-tools-will-need-people-to-drive-them thesis is respectable. But they are wrong in a way that will be visible in the 2030 workforce, and no lab's benchmark can argue with it.