Computer science enrollments: −12% new students, AI begins to empty out the classes

Computer Science Enrollments: −12% of New Students, AI Is Starting to Empty the Classrooms

🔎 The Entry Ladder Has Just Given Way

The figures published on October 29, 2026 by the National Student Clearinghouse Research Center are nothing like just another administrative report. They describe a generational shift: in fall 2026, computer science enrollments are declining by double digits in the United States, across all degree levels. The field that has drawn crowds for twenty years is emptying out — and fast.

This is no calendar fluke. 2025 had already recorded the first decline in two decades (−8.1% at the undergraduate level). Fall 2026 amplifies the trend, and Bloomberg sums up the mood in one line: computer science programs are losing their students while AI takes the jobs. AI Weekly made it one of its through-lines of the week.

Why now? Because families did the math before the universities did. Four years of study, tens of thousands of dollars in debt, only to land in a junior job market that models like GPT-5.5 or Claude Opus 4.7 are reshaping in real time. This report details the numbers, separates the signal from the noise — and asks the only question that matters: what is left to learn when AI codes better than a beginner?

The Essentials

• −12.4% in 4-year computer science bachelor's enrollments, −12.3% in community college, −10.4% at the graduate level in fall 2026 (NSC Research Center).
• Meanwhile, bachelor's degrees across all fields are growing (+0.6%) and health programs are booming (+6.2%): this isn't a crisis of higher education, it's a targeted loss of interest in computer science.
• Second consecutive year of decline after −8.1% in 2025, the first drop in ~20 years: roughly −20% of new students cumulatively over two academic years.
• The cause cited everywhere: AI is reducing junior developer positions. Unemployment among recent CS graduates already reached 7% in 2024 (New York Fed), ~3 points above the average.
• The phenomenon is global: UK (−9%), Australia, Quebec, Europe — all signals point in the same direction.
• Second-order risk: a third "tech talent drought," which history shows takes more than a decade to resolve (CIO, 2026).

Recommended Tools

The university is fading, AI-assisted self-learning is on the rise. For those who want to break into tech through the side door, here's our selection.

One cross-cutting piece of advice: whatever the tool, the portfolio remains the real résumé. Host your projects — Hostinger does the job for a few euros a month — and show code in production rather than certificates.

The numbers: a double-digit drop, across the board

In fall 2026, enrollment in computer and information science is falling by double digits at every degree level in the United States, while overall bachelor's enrollment grows. It's the combination of the two that makes the data damning: the exodus isn't general, it's targeted.

Source: NSC Research Center, preliminary 2026 data, via Higher Ed Dive.

The computer science bachelor's degree posts the biggest decline of any major, even as higher education as a whole is doing well. Community college (−12.3%) and graduate programs (−10.4%) follow the same pattern. Combined with the 2025 decline, the CS bachelor's degree shows roughly −20% in new enrollments over two academic years: a regime change, not a fluctuation.

The signal was actually already there in the spring. Persistence among CS students — those who return the following semester, then the next year — was the lowest of the ten most popular majors (National Student Clearinghouse, August 2026). Dropouts along the way preceded the enrollment slump. Departments first see those who were already seated walk out, before even noticing the absence of those who were supposed to arrive.

And it's not just an American phenomenon:

When four different education systems record the same collapse in the same year, you can no longer blame local quirks.

Structural, not cyclical: the calculus for families has changed

This decline doesn't look like a cyclical dip. Previous contractions — 2001, 2008 — followed the economy: demand for programmers returned with growth. Here, the trigger is different: it's the entry-level tasks themselves that are being automated.

Writing a function, writing tests, fixing a simple bug: models like GPT-5.5, Gemini 3.1 Pro or Claude Sonnet 4.6 now do this in a few seconds, with quality that's sufficient for many internal needs. The first floor of the career rocket — the one where you learned by producing simple code — is shrinking before our eyes.

Companies aren't hiding it. At Meta, using AI is becoming a condition for advancement, not an option: we detailed this shift in our article on Meta making AI mandatory for promotions. When a tool becomes an internal evaluation criterion, the message sent to students is unambiguous.

The movement goes far beyond one company. Amazon is pushing its engineers to measure AI adoption — sometimes to the point of "tokenmaxxing" the metrics. Meta, for its part, has moved to the next stage with the most massive AI restructuring in the tech sector: 8,000 layoffs. The direction is the same everywhere: produce more with fewer hands, starting with the cheapest ones.

Let's be honest about the macro economy: at this stage, AI's measured ROI remains low. But families don't vote on productivity audits: they vote on newspaper headlines, layoff announcements, and visible job postings. And there, the verdict is clear-cut.

A methodological clarification, because it matters: the author of the NSC data himself refuses to causally attribute the decline to AI and treats his figures as descriptive (VnExpress, 2026). That is the appropriate caution. But when British (BCS), Australian (ACS), and Quebec administrators spontaneously point to the same cause, the correlation stops looking like a statistical fluke.

The junior market is closing, and the ladder with it

Unemployment among young computer science graduates was already hovering around 7% in 2024 — and 7.8% for computer engineering — roughly three points above the overall rate (Federal Reserve Bank of New York, 2024 data, via Higher Ed Dive). Nothing suggests improvement since: hiring slowdowns and tech layoffs have continued throughout 2025-2026.

It needs to be called by its name: the ladder paradox. Today's seniors were yesterday's juniors. If AI absorbs the tasks that used to train juniors, and if companies stop hiring at the bottom of the ladder to optimize their costs over eighteen months, they are sawing off the branch they will be sitting on in eight years. The market isn't disappearing — it's being re-priced, and experience is worth more than ever.

Hence the coldly rational trade-off facing a high schooler in 2026: why pay for four years of study and take on debt for an uncertain first job, when three years of concrete AI-assisted projects can lead to the same portfolio? This question is no longer provocative. It has become the default question, and enrollment tables are the answer.

"Prompt engineering" vs. a CS degree: the debate is framed all wrong

Pitting prompt engineering against a computer science degree makes no sense. The former is a skill that can be absorbed in weeks; the latter builds technical judgment that pays dividends for decades. Reducing the debate to this head-to-head is like comparing a trim package to a chassis.

Prompt engineering as a job title is evaporating anyway, and that's logical: each generation of models — GPT-5.5, Claude Opus 4.7 (Adaptive), Gemini 3 Pro Deep Think — needs less and less hand-holding. The value has migrated to orchestration, evaluation, and architecture, not to elegantly wording a request.

What AI doesn't replace, on the other hand: arbitrating trade-offs, taking responsibility, securing systems, understanding a business context, navigating ten-year-old legacy code. In other words, precisely what a serious CS education aims for — algorithms, systems, networks, mathematics. The content isn't obsolete; the narrative is what broke.

My honest take: the real problem with the degree isn't pedagogical, it's narrative. Its ROI has collapsed for the average student aiming for a generic developer job — because that role is becoming scarce. For AI research, security, or infrastructure, it remains the best springboard I know of. For "a dev job in two years," the alternatives now win on the cost/time ratio. It needs to be said plainly, and universities must adjust their messaging accordingly.

Self-Taught Developers and Alternative Schools: The Side Door Is Opening

While computer science lecture halls empty out, alternative pathways are capturing part of the demand — at marginal cost, with AI as a personal tutor. The main door is narrowing, the side door is widening.

The most visible example: OpenAI Academy and its free courses to prepare for "the next era of work." When the industry leader trains people for free in the very skills it is looking for, the universities' historic monopoly on certification is publicly cracking.

The personal tutor, precisely. With models like DeepSeek V4 — whose Pro and Flash versions shake up the quality/price ratio — a self-taught learner can iterate without limits: generate, have it explained, have it corrected, start over. No teaching assistant has that kind of time, no matter how good their intentions. This is the first time in history that individualized tutoring in computer science is nearly free.

Add the proven structures — freeCodeCamp, CS50, The Odin Project — and you get a complete pathway for the price of web hosting. Recruiters, for their part, increasingly look at what you've built rather than what you're credentialed in.

One caveat, however: the alternatives lower the cost of entry, not the difficulty of the junior market. They change who can try their luck, not the number of spots available. Don't confuse a wider door with a less crowded room — that's the mistake many candidates are about to make.

What's Left to Learn When AI Writes Code?

When code generation becomes automated, value shifts to what AI doesn't do: framing, arbitrating, verifying, securing, taking responsibility. Here are the five skills that, in my view, will survive the decade:

• Architecture and system design: deciding what to build, and why, before writing a single line.
• Critical review of generated code: spotting security flaws, blind spots, and performance illusions in a model's output.
• Fundamentals (algorithms, networks, OS): this is what lets you understand why the AI gets it wrong — and therefore how to fix it.
• Domain knowledge: AI doesn't know what matters to your users; you do, if you make the effort to learn it.
• Agent orchestration: chaining GPT-5.3 Codex or Claude Sonnet 4.6 into reliable, measurable, and maintainable workflows.

The Two-Speed Rule

Two students use the same model. The first treats it as a shortcut: he copies, it works, he moves on. The second treats it as a demanding tutor: he has it generate, then explains every line, then rewrites it from memory.

Eighteen months later, the first is dependent; the second is formidably competent. The tool amplifies who you are; it doesn't replace you. That's the only lesson about learning worth remembering from all this news.

Second-Order Effects: The Third "Talent Drought" Looms

If the current decline persists through three or four more academic years, the economy will lack experienced computer scientists in the early 2030s — precisely when AI-driven demand will peak. That is the scenario outlined by CIO in 2026: the two previous contractions of the tech talent pipeline each took more than a decade to resolve.

This contraction has an unprecedented feature: it strikes both the pipeline's entry point (enrollments) and the senior-engineer factory (training itself, weakened by the drop in persistence observed as early as spring 2026). Previous shortages hit demand; this one hits supply at the root.

Read this signal the other way around, too: the 2026–2030 cohorts who persevere will face reduced competition. Scarcity pays — exactly what happened to graduates who entered the market just after 2012, arriving on cleared ground before the next rush. Tech counter-cyclical entrants have historically fared well.

That leaves institutions with a responsibility: if universities want to reverse the curve, they will have to offer an "AI-native" curriculum — fundamentals + orchestration + evaluation — rather than defending a program perceived as dated. Otherwise, the decline will continue, and it will not be cyclical.

❌ Common Mistakes

Mistake 1: Reading this decline as the death of the developer profession

What's wrong: confusing the compression at the bottom of the ladder with the disappearance of the profession. Demand is shifting toward more experienced and specialized profiles. The solution: aim for depth (systems, security, data) rather than the volume of code produced.

Mistake 2: Sacrificing the fundamentals in favor of tools

What's wrong: tools change every six months — Grok 4.1, GLM-5.1, Kimi K2.6 replace one another. Algorithms, networking, and systems, on the other hand, compound. The solution: roughly 70% fundamentals, 30% tools, and reassess each semester.

Mistake 3: Confusing correlation with causation

What's wrong: the author of the data refuses to attribute the decline to AI; the figures are descriptive. The post-COVID correction, salaries perceived as less attractive, and the highly publicized layoffs also play a role. The solution: present AI as a probable accelerator, not as a proven sole cause.

Mistake 4: Declaring the degree useless too quickly

What's wrong: for research, security, infrastructure, and access to certain foreign job markets, the degree remains a very real entry ticket. The solution: evaluate the degree based on career goals, not based on press headlines.

❓ Frequently Asked Questions

Should you still pursue a computer science degree in 2026?

Yes, if you're aiming for AI, security, systems, or research: the degree remains the best springboard. No, if your goal is a generalist web developer position: alternative pathways cost less and get you to a portfolio faster. In any case, learn the fundamentals — that's what withstands automation.

Is AI the sole cause of this decline?

No. The researcher behind the data rejects strict causality: the post-COVID correction, the highly publicized layoffs, and salaries perceived as less attractive also play a role. But AI is the most plausible accelerant: it hits precisely the entry-level job openings that motivated enrollment in the first place.

Does prompt engineering replace a CS degree?

No. Prompt engineering is a skill, not a profession: recent models need it less and less. What's lacking are people capable of architecting, auditing, and making reliable what AI produces — and that requires exactly the foundational training that a degree provides.

Which fields benefit from this loss of interest in computer science?

Healthcare, first of all: +6.2% in bachelor's enrollment in fall 2026, the exact mirror image of computer science. Data and applied mathematics programs are also holding up better. Students are arbitraging toward careers perceived as sheltered from automation — a debatable judgment, but one that is shaping real choices.

Will companies regret having cut juniors?

Probably, within five to eight years. The seniors of 2032 are the juniors we aren't hiring in 2026. The two previous contractions of the tech pipeline took more than ten years to work themselves out; this one could be more brutal because it strikes both the entry point of the pipeline and the training itself.

✅ Conclusion

IT isn't dying: its entry ladder — the junior position — is crumbling, and the 2026 generation is the first to have internalized this new deal before the institutions did. Degree or not, the winning strategy remains the same: solid fundamentals, real projects, AI as a lever — and an eye on our new AI tools so you don't end up learning with a ten-year delay.