This one started with a conversation, not a slide deck. My friend David has been through a tech cycle or two himself, and he asked me to go on a journey with him: sit down and actually think through the parallel between the dot-com crash and the AI era we're in now, rather than just repeating the version everyone already has in their head. That version usually stops at the vibe check: the fear rhymes, so the outcome must too. That's not enough. If we're going to argue this is a cycle and not a collapse, it's worth walking through what actually happened last time, with real numbers, and being honest about where the parallel holds and where it's still a guess.
The Parallel, With Numbers Attached
Start with what's actually comparable. After the dot-com bubble burst in 2000, the fear was outsourcing: American programming jobs moving to Bangalore and cities like it, a story the press told constantly. Layoffs across the tech sector dominated headlines for years. "Programming is dead" wasn't a fringe opinion. It was a mainstream one, repeated by people who should have known better.
The numbers backed the fear up, at least for a while. New computer science majors at U.S. universities dropped by half between 2000 and 2007, from roughly 15,958 to 7,915 a year, according to Computing Research Association Taulbee Survey data. Communications of the ACM later put a finer point on it: enrollments hit half their 2000 level by 2005, then sat flat through 2007. That's not a soft correction. That's a field that lost half its incoming students in five years.
Now look at today. Undergraduate enrollment in computer and information sciences at four-year colleges fell 8.1 percent in fall 2025, from about 659,700 students to 606,100, according to the National Student Clearinghouse. A Stanford economist working from a dataset of more than 50 million course sections going back to 1996 found that computer science's share of total enrollment fell for the first time in roughly two decades. The Computing Research Association's own 2025 Taulbee Survey shows new CS bachelor's enrollments down nearly 13 percent, master's down over 10 percent, and doctoral down 15 percent. At Duke, introductory CS enrollment dropped 20 percent in a single year. At Princeton, the number of students declaring a computer science degree fell from 150 to 117 to 74 over two admissions cycles.
The fear is the same shape: automation standing in for outsourcing, "coding is dead" standing in for "programming is dead," layoffs dominating the coverage again, this time blamed on AI instead of offshoring. What's different is the scale, at least so far. An 8 percent national drop in one year is a real signal. It is not yet a 50 percent collapse over five years. We're early in this cycle, not deep into it, whatever the parallel slides imply.
What Actually Happened After the Last Crash
Here's the part that gets flattened into a slogan too often: computer science didn't just "come back." It came back unevenly, slowly, later than people expected, and different than it left.
New CS enrollments stayed essentially flat from 2005 through 2007. The first real increase didn't show up until the 2007-08 academic year, when new majors rose 8 percent, the first jump since 2002. That's seven full years from bust to first sign of recovery. A student who chose computer science in 2003, betting the field was cyclical and would come back, spent most of their undergraduate years in a genuinely bad job market before it turned.
The recovery also wasn't a return to the 2000 landscape. By 2007-08, the industry pulling in that new cohort of majors already looked different from the one that had cratered in 2001. Web applications had matured into permanent infrastructure. Broadband and always-on connectivity had opened up a security surface that barely existed in 1999. Cybersecurity grew directly out of that gap, into a discipline with its own degree programs, hiring pipelines, and salary premiums. It wasn't the whole story of the recovery, but it was a real and durable piece of it, and that's why "cybersecurity eventually surged" earns its place in any honest version of this comparison.
The Mechanism, Not Just the Mood
The parallel is worth taking seriously not because the fears happen to match, but because both crashes run on the same mechanism. A technology shift outpaces the market's ability to absorb it. Panic sets in faster than the labor market can adjust. Enrollment drops in response to a snapshot of bad news rather than a five-year forecast. The correction overshoots. Employers who need CS talent in 2029 or 2030 will be drawing from a smaller pool than they'd like, because a chunk of a whole generation of capable students opted out during the scary years, based on incomplete information. That's what happened after 2000. There's no obvious reason it wouldn't happen again.
The current data backs up that "incomplete information" part. The Stanford economist behind the 2025-26 numbers has been explicit that his data is descriptive, not causal, and that he doesn't pin the decline on AI alone. He points to a weaker junior-developer job market and general doubt about the value of a CS degree as separate, compounding factors. Some of what's driving students away right now isn't really about AI writing code. It's a real hiring slowdown, mixed with an AI narrative that's louder than the underlying data currently supports. That combination, real signal plus amplified narrative, is close to what happened with outsourcing in the early 2000s too. Offshoring was real. It also got blamed for more of the crash than it probably caused.
What This Means If You're Deciding Right Now
If the historical timeline holds even loosely, here's the shape to expect: a few more years of declining or flat enrollment while the job market and the AI narrative both sort themselves out, a first sign of recovery around 2028 or 2029, roughly seven to nine years after this decline started, and a graduating cohort by 2030 walking into a field that looks meaningfully different from the one that scared their predecessors away.
Here's my own read on where this goes. The AI era, like every wave before it, will plateau and then dip. The easy gains get captured first, the hype outruns the actual capability, and reality sets back in. That's exactly the point where industry realizes it doesn't have enough people who know how to build, secure, and run what got created during the boom. That's when the surge in demand for skilled workers hits. Right now, plenty of students are reading the fear correctly as fear and quitting the field because of it. The ones who stick it out through the plateau are the ones who'll be standing at the front of the line when that demand surge arrives, the same way the 2007-08 cohort was standing there when the dot-com correction finally ran its course.
The specifics of that difference matter more than the comfort phrase "technology careers evolve." Last time, the field widened to absorb the internet's permanent connectivity, and cybersecurity was the clearest new specialty to come out of it. This time, the gap is being opened by generative AI and autonomous systems, and the specialties forming around it are AI and machine learning development, data science, security work aimed specifically at model pipelines and agent behavior, and software engineering practiced with AI tools instead of against them. None of that works without the same foundation it always required: real computer science fundamentals, the kind that don't go stale when the specialty around them changes.
The Fundamental That Actually Matters
If I had to name the one skill underneath all of those fundamentals, it's critical thinking, not memorization, and not the kind of testing we've built our education system around for the last few decades. Standardized testing rewards getting to the right answer fast. It doesn't reward knowing why the answer is right, or being able to tell a good answer from a confident-sounding wrong one. That distinction used to be a nice-to-have. With AI in the picture, it's the whole job.
Any student can now get an AI system to produce code, an analysis, or a design in seconds. What separates a valuable computer scientist from a replaceable one isn't who can produce that output fastest, it's who can look at it and tell the good from the bad, and explain why. That's not a skill memorization builds, and it's not a skill multiple-choice testing measures. It's a skill built through active learning and project-based work, the kind where a student has to defend a decision, debug a failure, and justify a tradeoff in front of someone who'll push back. Programs that lean into that kind of teaching are the ones that will actually produce the graduates who show up ready for what's coming. Programs that keep optimizing for standardized test performance are going to keep producing students who can operate AI tools without being able to judge what those tools hand back to them.
History's advice to the 2003 cohort was simple, even if most of them didn't hear it in time: stay in, keep building fundamentals, and pick up whatever specialty is emerging instead of waiting out the fear. The students who do that this decade, and who build the critical thinking to go with it, are the ones who end up graduating directly into the shortage everyone else was too scared to prepare for.
That's what David and I landed on, and it's worth thinking through yourself, whether you're a student, a parent, or just someone who's watched this movie before.
