At a glance
Stanford's computer science department remains ranked first in the country, yet bachelor's degrees fell nearly 14 percent in a year and entry-level hiring in Silicon Valley has thinned sharply. What the payroll data, the AI Index and the people doing the hiring actually show.
The Rankings Still Say One. The Queue Says Something Else.
For as long as most applicants can remember, U.S. News has placed Stanford's computer science department at or near the top of its graduate rankings, and the 2026 edition keeps it there — joint first among American computer science schools. So the most interesting number out of Palo Alto this year isn't a rank.
It's 70.
That's roughly how many fewer bachelor's degrees in computer science Stanford awarded in 2026 than the year before, a fall of nearly 14 percent in a single cycle. The major itself shrank earlier. In autumn 2023, 865 undergraduates had declared computer science. By autumn 2025, the figure was 639. Still triple the size of economics, the next largest major on campus. Still the biggest department by enrolment. Just no longer growing, for the first time since the dot-com crash.
Why should anyone outside California care? Because for twenty years Stanford computer science has been the reference implementation of a bargain that got exported everywhere: learn to code, get hired, get rich. When the reference implementation wobbles, the wobble travels. Britain's IT labour market is soft, Indian engineering graduates are competing for the same shrinking pool of junior roles, and recruiters in Berlin and Singapore read Bay Area hiring data the way farmers read frost warnings.
What the Department Actually Is
Computer science became a department at Stanford in 1965, inside the School of Humanities and Sciences, and for its first two decades it taught graduate students only. It moved into the School of Engineering in 1985, and the undergraduate major opened a year later.
The department's shape owes a great deal to Frederick Terman, the engineering dean and later provost who built what he called steeples of excellence rather than spreading money evenly across fields. Terman didn't try to construct a computer in the electronics labs the way rival universities did. He recruited the mathematician George Forsythe to run a small computer science division inside the mathematics department and let it grow into an independent department with a faculty assembled largely from the East Coast. He also leased university land to technology companies, a decision that did more to create the industrial cluster around the campus than any curriculum ever did. The phrase Silicon Valley was coined by a journalist and Terman never took to it, but by the time he died in 1982 the wire services were calling him the father of the electronics revolution.
The scale today is bracing. Sixty-four full-time faculty. About a thousand graduate students in computer science, more than 500 of them full-time master's candidates, with a PhD student-to-faculty ratio of 5.7 to 1. Full-time tuition listed at $65,865 a year, an application deadline of December 2 and a $125 application fee. Stanford students, staff and faculty have founded well over a thousand companies in the Valley over the decades, ventures that by various measures account for somewhere between 30 and 60 percent of employment in the region's high-tech firms. And the money still arrives: one analysis of federal earnings data puts median pay for Stanford's computer science bachelor's graduates at about $215,000 four years after they enrol, among the highest of any program in the country.
None of that has changed. What changed is the door on the other side.
Here's the Catch: The Bottom Rung Came Off
The strongest evidence about what happened to entry-level technical work isn't a survey or an anecdote. It's payroll data.
Economists at the Stanford Digital Economy Lab — Erik Brynjolfsson, Bharat Chandar and Ruyu Chen — used high-frequency payroll records covering millions of American workers and found that employment for people aged 22 to 25 in occupations heavily exposed to AI now sits about 19 percent below where it would be had it kept pace with similarly aged workers in less exposed jobs. When they first published the finding in 2025, the gap was 15 percent. It has widened, not closed. Experienced workers show no comparable gap.
Two details in that working paper matter more than the headline. First, the adjustment is happening through hiring, not firing: firms are not sacking young engineers in droves, they are declining to bring them in. Second, the declines cluster in occupations where AI substitutes for tasks, not where it assists. The authors are careful to call their findings descriptive, not causal, and they note the gap narrows when education is controlled for. Their explanation is intuitive anyway. Generative AI is very good at reproducing codified knowledge — the textbook material that formal education is built on — and much worse at tacit knowledge, the judgment that accumulates through practice. A new graduate has a lot of the first and very little of the second.
The Census Bureau has reached a compatible conclusion from a different dataset. Economists Lee Tucker and Lawrence Warren found that graduates entering highly AI-exposed fields after ChatGPT's release were less likely to find work and more likely to start in lower-paid roles. Tucker's summary is blunt: the changes are on the order of what previous research suggests happens in a large recession.
What the People Doing the Hiring Say
Jan Liphardt, an associate professor of bioengineering at Stanford, told the Los Angeles Times that the university's computer science graduates were struggling to find entry-level jobs at the most prominent tech brands. His reaction to his own sentence: "I think that's crazy."
The arithmetic behind that shift is not mysterious. Nenad Medvidovic, a computer science professor at the University of Southern California, described the new staffing model plainly: where a project once needed ten engineers, it now needs two skilled ones plus a large-language-model agent. Amr Awadallah, chief executive of the Palo Alto startup Vectara, put it without polish. "We don't need the junior developers anymore," he said. "The AI now can code better than the average junior developer that comes out of the best schools out there." Dario Amodei of Anthropic has said Claude writes the large majority of code in some of his company's products and has predicted that a substantial share of entry-level white-collar work could be displaced within five years.
Students describe a market that has split rather than collapsed. A narrow band of graduates — the ones classmates call cracked engineers, with shipped products, research output and fluent use of AI tooling on their CVs — still get multiple offers. Everyone else competes for a thinner set of openings. "There's definitely a very dreary mood on campus," one recent graduate told the Times, asking not to be named.
Photo by Brecht Corbeel on Unsplash
The Fifth-Year Master's, and Other Ways of Not Leaving
The most visible response on campus isn't a protest or a petition. It's re-enrolment. A large cohort of frustrated graduates has opted for a fifth-year master's, buying another recruiting cycle and another set of credentials. "Half of my friends are still on campus," the same graduate said. Master's enrolment in the department has moved the opposite way to the undergraduate major, rising even as the bachelor's pipeline thinned.
Whether that's a shelter or a delay is genuinely unclear, and the cost of finding out is not trivial in a rental market like Palo Alto's. A Stanford master's is an expensive way to postpone a job search.
Watch the curriculum, though, not just the headcount. MIT split its computing degrees years ago and its newer Artificial Intelligence and Decision Making major, launched in 2022, has grown into one of the larger undergraduate programs on that campus while the classic Computer Science and Engineering degree slipped. Stanford's chair, Mehran Sahami, has resisted the idea that a bad hiring cycle should force a rebuild, telling MarketWatch that it would be "short-sighted to just say there's been a round of layoffs, so suddenly everything needs to change at universities." Nationally, enrolment in the major fell roughly 8 percent in autumn 2025 after nearly two decades of growth, according to an analysis of National Student Clearinghouse data, and the Hoover Institution economist Jacob Light found the subject's share of course-taking declined in 2025–26 for the first time in about twenty years — while explicitly declining to call AI the cause.
The Visa Variable Silicon Valley Can't Ignore
Roughly two-thirds of H-1B positions are computer-related. Amazon alone had more than 10,000 H-1B approvals in the first half of 2025; Microsoft and Meta each exceeded 5,000. For an international graduate of a Stanford master's program, the path from diploma to desk runs through a programme that has spent two years in court.
In September 2025 the Trump administration attached a $100,000 payment to new H-1B petitions, against a previous total cost of roughly $2,000 to $5,000 per worker. In June 2026 a federal judge in Boston, Leo Sorokin, vacated the fee as an unlawful tax that only Congress could impose, and the First Circuit declined to stay that ruling while the government appeals. A second judge in California blocked enforcement again on procedural grounds at the end of September 2026. Separately, the Department of Homeland Security has proposed a $103,265 fee on cap-subject petitions through formal rulemaking; the comment period closed in September and the fee is not in effect. The litigation is unresolved in the way that matters to a recruiter deciding next quarter's headcount.
There is a second front. About 419,000 people were working in the United States on Optional Practical Training in 2024, the programme that lets foreign graduates work for one to three years on a student visa. The administration has weighed attaching a $100,000 charge to it and is drafting a broader rewrite of the rules. OPT is the route by which most international graduates of American universities actually enter the tech workforce; a fee at that level would hit the firms that hire from top campuses more directly than the original H-1B levy did.
Grading the Hype
Four competing stories are being told about Stanford computer science right now. They are not equally true.
| Claim | Grade | What the evidence supports |
|---|---|---|
| It is still the strongest computer science department in the country | A− | Joint first in the national rankings, 64 full-time faculty, and no peer has matched its record of spawning companies. Rankings measure faculty reputation, though, not your offer letter. |
| AI has eliminated junior software jobs | C | Employment for 22-to-25-year-old developers is down roughly a fifth relative to peers. That is a severe thinning, not an elimination. New graduates are still being hired — fewer of them, at a higher bar. |
| The degree is now worthless | F | Median earnings four years out sit near $215,000. Demand thinned at the entry point, not at the top. |
| Universities must rebuild the curriculum immediately | C+ | Sahami's scepticism about reacting to one hiring cycle has real force. MIT's reallocation suggests the on-ramp has moved regardless. |
One more piece of context that rarely makes the panic pieces. The Stanford AI Index, the field's annual scorecard, reports that global corporate investment in AI hit $581.6 billion in 2025, up about 130 percent in a year, and that organisational adoption reached 88 percent. By late 2026 the index team put year-to-date corporate AI investment above $1 trillion, already 46 percent beyond all of 2025. The same update noted something the simple story misses: employment declines had begun deepening even in occupations where AI is supposed to support rather than replace people, including marketing and IT management roles. Capital is pouring in faster than anyone can measure what it displaces.
What a 2026 Graduate Can Still Control
Xiaoyin Qu, a Stanford dropout and former Meta product manager who has founded two venture-backed companies, has watched the split from the hiring side. Degrees from Stanford and MIT land in her inbox attached to messages asking whether she knows anyone hiring interns. The graduates fielding premium offers, she says, share a habit rather than a pedigree: they ship things in public and can demonstrate what they built.
"The credential filter is weakening. Proof of execution is replacing pedigree," she told interviewers earlier this year. Her advice to students is a four-verb sequence — build, ship, compete, publish — and she points to a nineteen-year-old who won a hackathon run by xAI and was hired on the spot.
That is an extreme case, and it should not be generalised into a rule. But it lines up with what the labour economists prescribe. If AI has devalued codified knowledge, then the thing that still compounds is tacit knowledge: judgment built by shipping software that real users touched, by debugging systems nobody handed you a spec for, by working alongside people who have run production. Internships, open-source contributions and research placements do that work. A fifth-year master's does it only if the year contains something other than coursework.
And the specific technical skill that employers now screen for is not writing code from a blank file. It's directing, reviewing and correcting machine output at a level a junior developer couldn't reach two years ago. Every hiring manager quoted in this piece said some version of the same thing: the work didn't vanish, it moved up a level.
Photo by Brecht Corbeel on Unsplash
The Chair's Argument
Sahami's case for calm rests on history, and the history is genuinely awkward for the doom narrative. When computer science enrolment fell in the early 2000s after the dot-com collapse, the students who ignored the signal and majored anyway graduated around 2006 and 2007 — into the founding years of Facebook and the steepest growth phase Google would ever have.
His own reading of the current moment is that firms will take time to absorb what he calls the AI dividend, the productivity gain from engineers working with these tools, and that the job will re-form around people who can use the tools better rather than around people who can out-type them.
"If you imagine someone who was going into college in the year 2001, 2002 — when computer science enrolments were dropping — that actually would have been a great time to major in computer science," he said. Whether 2026 is that year or merely rhymes with it is the question every seventeen-year-old weighing the major, and every graduate weighing a fifth year, is now trying to answer with incomplete data. The local reporting on the degree decline makes the same point less comfortably: a downturn in the pipeline isn't automatically a verdict on the field. Sometimes it's a lag. Sometimes it's the market telling you where the ladder moved.
- h-1b-visa
- stanford-computer-science
- silicon-valley-hiring
- computer-science-enrolment
- ai-and-entry-level-jobs

