The 2031 Crisis in Higher Education

American higher education in 2025 was a patient with a pre-existing condition. Several, actually.

The demographic cliff had been forecast for over a decade. The number of 18-year-old high school graduates peaked at roughly 3.9 million in 2025 and then began the descent that the Western Interstate Commission for Higher Education had been warning about since the mid-2010s: a projected 9% decline in the high school graduate population between 2026 and 2031³, with the steepest drops concentrated in the Midwest and Northeast. By 2041, Illinois was predicted to have 32% fewer graduates; California: 29%; New York: 27% fewer.⁴ Nathan Grawe, the Carleton College economist who had spent a decade modeling this, took no satisfaction in being right.

The sector had known the cliff was coming and hadn't sufficiently prepared. Instead, it had engaged in a decade-long arms race of competitive discounting. By the 2024–25 academic year, the average tuition discount rate at private nonprofit institutions had reached a record 56.3% for first-time, full-time undergraduates.⁵ Although the sticker price at a private college was $45,000 a year, the average student actually paid less than half. The schools were collecting roughly 44 cents on every dollar of listed tuition, and the most vulnerable ones far less. Net tuition revenue was still rising, barely, by about 1.4% in real terms. But the math only worked if enrollment held. Each incoming class was recruited at a deeper discount than the last, and the strategy depended entirely on a supply of students that demography had already promised to take away. It was a treadmill, and the belt was accelerating.

Already in 2024, more than half of private universities rated by S&P Global generated operating deficits.⁶ Student loan debt stood at $1.833 trillion across 42.8 million borrowers. Federal student loan delinquency had climbed over 10% by late 2025, with 20% of borrowers behind on payments. Over 100 colleges had closed or merged in the prior eight years.⁷ The share of high school graduates immediately enrolling in college had dropped from 70% to 62% over the prior decade.⁸ And in the final weeks of 2025, all three major credit rating agencies issued negative outlooks for the higher education sector in rapid succession: Moody's on November 21, S&P Global on December 3, and Fitch on December 5.⁹

Meanwhile, the employers on the receiving end of the pipeline were quietly dismantling the requirement that had funneled students into the system in the first place. By 2023, Google, Apple, IBM, Bank of America, Walmart, Accenture, and Delta had all dropped bachelor's degree requirements for significant categories of roles.¹⁰ By 2025, a quarter of employers surveyed planned to eliminate degree requirements for some positions by year's end, and 70% of hiring managers said they prioritized relevant experience over a diploma.¹¹ The trend was still early, still unevenly applied, and still easy to dismiss as a tight-labor-market anomaly. But it revealed something the universities preferred not to examine: that the employers who had spent decades requiring bachelor degrees were discovering they didn't actually need them.

The 2024-2025 period showed the early signs of contraction. Birmingham-Southern College, which was 168 years old and once home to six Rhodes Scholars, had bled its endowment from $122 million to $26 million. Enrollment had collapsed from 1,500 to 731 before its final commencement.¹² Sixteen more nonprofit institutions closed in 2025¹³: Fontbonne University in St. Louis (founded 1923), Northland College in Wisconsin (founded 1892), Eastern Nazarene College near Boston, and Limestone University in South Carolina (which owed 281 students nearly $400,000 in refunds when the doors shut).¹⁴ Martin University, an HBCU in Indianapolis, paused operations in December 2025 and closed in January 2026¹⁵ — early enough to be written off as an outlier, late enough to be recognized as a harbinger.

The question, even then, was no longer whether the system was contracting. It was whether the contraction would be orderly.

Penn State's board, meanwhile, approved the closure of seven of its nineteen Commonwealth Campuses, the satellite locations that had served as on-ramps to the Penn State degree.¹⁶ The board voted on the closures in the same session it reaffirmed the $700 million Beaver Stadium renovation. The brand would be preserved. The access mission would not.

Each followed the arc Hemingway described for bankruptcy: gradually, and then suddenly.

But in 2025 and early 2026, the closures were still understood as isolated failures: bad management, bad luck, niche institutions with niche problems. The system, everyone agreed, was under pressure. The system was not in crisis.

But the system was in crisis.

By early 2026, the fracture was already visible in the 2025 data itself. When the National Student Clearinghouse released its final fall enrollment numbers in January, total undergraduate enrollment had recovered to 16.2 million, within 400,000 of the 2019 peak. The headline read as recovery. The composition read as warning. Community college enrollment rose 3.0%. Public four-year rose 1.4%. Private nonprofit four-year declined 1.6%. For-profit four-year declined 2.0%. The aggregate was climbing. The institutions most exposed to the coming storm were already contracting, inside what was otherwise a recovery year.

The first tremor that should have rewritten everyone's assumptions arrived in the fall of 2026, and almost nobody recognized it for what it was.

Computer science enrollment — the hottest major in American higher education for a decade, the field parents pushed their children toward, the degree that almost guaranteed an eventual six-figure salary — was declining.

Not at a handful of struggling schools. Across every institution type and award level. Undergraduate CS fell 3.6%. Graduate CS fell 14.0%.¹⁷

This was counterintuitive. AI was the most transformative technology since the internet. Surely the people who built AI tools would thrive? But the agentic AI systems that emerged in early 2026 (Claude Code, Codex, Cursor, and their successors) did not simply automate tasks. They restructured the economics of software development. A team of twenty became a team of five. Amazon shed 16,000 jobs in January. Dell cut 11,000 in March. Meta 8,000 in May. Even Microsoft was now offering voluntary retirement to 7% of its workforce. By fall 2026, the pattern was accelerating.

In the financial markets, the iShares Software ETF fell roughly 24%.

Challenger, Gray & Christmas reported in March of 2026 that AI was the single most-cited reason for layoffs in the United States, accounting for 25% of all job cuts that month. Technology alone shed 52,050 workers in Q1 2026, up 40% year-over-year. The man whose firm tracked the data put it plainly: "Companies are shifting budgets toward AI investments at the expense of jobs."¹⁸ But these early signs were often ignored or dismissed as the mere 'AI-whitewashing' of regular layoffs.

But for the parents of a high school junior watching these layoffs, the calculation changed overnight. If AI could write better code than most mid-level developers, what exactly were they buying for $200,000?

The establishment treated the CS enrollment dip the way Wall Street treated subprime mortgage defaults in early 2007: contained. A sector-specific correction. Technology adjusting to new tools. "Students are just recalibrating," a dean at a top engineering school told the Chronicle of Higher Education in February 2027. "The fundamentals of a CS education remain strong."

They didn't realize it was the canary. If AI could displace computer science graduates, which were the most technically sophisticated products the university system produced, it could also displace anyone whose work was primarily cognitive. And that was not a niche. That was the entire white-collar economy, which represented roughly 70% of the jobs university graduates had historically filled.

A friend of ours is a senior product manager at a mid-sized SaaS company in Dallas with a Baylor MBA and twelve years of experience. He was laid off in March 2027. His entire product team was restructured around AI tools. He spent four months applying for roles before accepting a position at 60% of his previous salary. "For twenty years the degree opened every door I needed," he said. "Now the doors are gone. I'm competing against agents that never sleep, never ask for raises, and keep getting smarter every week."

By the fall of 2027, the broader impact was technically measurable, but still easy to misread. Freshman enrollment at non-elite private institutions was down roughly 8% from 2025 levels. At regional publics, 6%. The numbers were bad, but within the range demography alone could explain, and administrators attributed them accordingly. Meanwhile, applications to the top 50 national universities increased. Harvard, Princeton, and Yale hit record numbers. The bifurcation had begun. Almost no one recognized it yet. The flight-to-quality that economist Richard Vedder had described for years went into overdrive.

The universities that saw what was coming did what universities always do: they launched new programs.

Lipscomb University in Nashville, TN went all in. By 2024, it had launched Nashville's first graduate program in applied AI — a Master of Science and certificate track designed to equip working professionals with AI fluency. In 2025 campus-wide enterprise AI access followed. An AI committee. An official usage policy. AI bots embedded in the education school's courses. Lipscomb's provost said what every provost was thinking: "Our mission is to prepare students for a world that doesn't yet fully exist."

The University of South Florida enrolled 3,000 students in a new AI and cybersecurity college. Columbia, USC, Northwestern, and the University at Buffalo all launched AI-specific degree programs. MIT's AI and Decision-Making major became the second-largest on campus. Across the country, universities were doing what the adaptation blueprint said they should: pivoting toward the technology, building curricula around it, positioning their graduates to ride the wave rather than be drowned by it.

It was the right response to the wrong question.

The programs were primarily built around "AI literacy," which is the ability to work productively with AI tools, to understand their capabilities and limitations, to integrate them into professional workflows. The question was whether this would be a durable, high-value skill that justified years of study and tens of thousands of dollars in tuition. For a narrow window, roughly from 2026 to 2029, it was. Employers wanted people who understood prompt engineering, model selection, retrieval-augmented generation, agentic workflows. The universities that moved fast captured that demand.

But AI literacy had the shelf life of a transitional skill, not a professional discipline. The analogy was not learning to practice medicine. The analogy was learning to type.

In 1985, community colleges across America offered typing courses. They were practical, vocational, and genuinely useful because most adults did not know how to use a keyboard. By 1995, every teenager in the country could type, because they had grown up with computers in their homes and schools. The skill was absorbed into the ambient competence of the population. The courses disappeared, not because typing stopped mattering, but because it stopped requiring formal instruction.

AI fluency followed the same trajectory, faster. A child who began using Claude or its successors at age thirteen grew up building projects with AI assistants. They learned to prompt naturally the way earlier generations learned to Google, and this new generation did not need a $160,000 degree to become "AI literate." They already were. By 2029, most high schools had absorbed AI tools into everyday coursework the way they'd once absorbed graphing calculators. AI was no longer a subject to study but infrastructure to use. Students were using AI to research, write, and work through problem sets before they'd filled out a college application. By 2030, the first cohort of AI natives were entering the workforce without having taken a single college course in applied AI. They didn't need to. Asking them to pay for AI literacy training was like asking a native English speaker to pay for English lessons.

For older workers retraining mid-career, the value of structured AI education lasted somewhat longer. But even here, the economics were fatal to the university model. An eight-week bootcamp, a $500 online certificate, or a corporate training program could all deliver AI fluency at a fraction of the cost of a graduate degree, on a timeline that matched the urgency of career transition. The market for AI upskilling was real, but it was a market for speed and affordability, not for the slow, expensive, credit-hour-denominated education that universities were structurally designed to provide.

The deeper problem, though, was not that universities were teaching the wrong skill. It was that they were training students for a labor market whose entry points were being eliminated.

On March 13, 2026, at Palantir's AIPCON conference, company architect Chad Wahlquist described the transformation of military intelligence targeting using the Maven Smart System. Work that once required roughly 2,000 intelligence officers (identifying targets, analyzing data, developing courses of action) now required about 20, and they were doing it faster.¹⁹

The ratio was not a prediction. It was an operational fact, reported during an active U.S. military campaign. In that same month, job postings for entry-level roles on Handshake had fallen 16% compared to the prior year, while applications per role surged 26%.²⁰ Fifty-eight percent of 2024 and 2025 college graduates were still looking for their first job.²¹

By early 2029, that ratio — not the specific number, but the structure of labor compression — was repeating across every white-collar field the university system fed. The entry-level financial analyst who built models in Excel: replaced by a senior analyst with a team of AI agents. The junior associate who spent two years doing document review: replaced by a platform. The first-year management consultant who assembled slide decks from data: redundant. The entry-level marketing coordinator who wrote copy and managed campaigns: automated.²²

Although no corporate setting had replicated Maven's 100:1 compression, ratios of 5:1 and 10:1 were becoming routine. A McKinsey engagement that once staffed eight analysts and two partners now staffed two analysts and two partners. A mid-size accounting firm that once hired twelve first-year associates hired three. In each case, the senior people survived. They had judgment, relationships, institutional knowledge, and the capacity to evaluate AI output rather than merely produce it. The work itself didn't disappear. The bottom of the pyramid did.

This headline, when it landed, should have ended the debate. It didn't, because by October 2028 there were too many crises to track. But it mattered more than almost anything else that happened that year, because of what it revealed about the internal economics of universities.

Professional schools (law, business, accounting) were not just programs. They were profit centers that charged higher tuition and offered lower discount rates to generate the surplus revenue needed to sustain the rest of the university ecosystem. When these programs shifted from surplus-generating to running deficits, the entire university budget came under simultaneous pressure. There was no reserve. There was no buffer. The cross-subsidy had been the load-bearing wall, and it collapsed.

Law school, with exquisite irony, had been booming. LSAC data showed applicants up 18% for the 2025 cycle²³ and 33% year-over-year in early 2026 data.²⁴ First-time LSAT takers surged: up 18% in August, 24% in September, and 16% in October.²⁵ Mike Spivey, a veteran admissions consultant, said he had never seen a cycle start so strong in 26 years of watching the data.²⁶ The reasons were familiar: when the economy frightens people, they flee to professional school. It was shelter-seeking behavior. The 2008 script, run again.

But it wasn't shelter. It was a trap. The signals that law school deans and the general public weren't paying attention were found in the VC markets. Legal AI investment topped $5 billion in 2025²⁷ and another $10 billion by the end of 2026. Startups like Legora had raised $800 million in a matter of 10 months.²⁸ In May of 2026, Anthropic announced Claude Legal Solutions, which gave lawyers native Claude integrations into 20 software platforms across 12 different practice types.²⁹

By 2028, AI tools could perform document review, contract analysis, due diligence, and regulatory compliance at a level that had required teams of junior associates billing $400 an hour. Large firms flattened their pyramids. Regional firms discovered that a solo attorney with an AI subscription could handle a caseload that previously required three associates and a paralegal. The class of 2028 graduated $150,000 to $250,000 in debt into a market that no longer valued what they'd purchased. The application surge of 2026 became, by 2029, a violent reversal. At least a dozen law schools closed or announced closure. Another thirty operated below sustainable enrollment.

A young woman we know in the top 15% of her class at a well-regarded (but not top-14) law school had her offer rescinded two weeks before her May 2029 start date. The firm had adopted an AI contract review platform and didn't need the associate class it had planned. She had $210,000 in debt. She took a compliance role at a bank at roughly half the salary she'd expected. "I did everything right," she told a mutual friend. "Every single thing they told me to do. And it didn't matter."

Accounting was hit even harder. The Big Four had been quietly reducing entry-level hiring since 2025, as AI automated audit procedures, tax preparation, and advisory analysis. PwC deployed an internal AI platform that could complete a standard audit workpaper in minutes rather than the hours it took a first-year associate. By 2028, Deloitte's campus recruiting events, which were once the lifeblood of accounting programs at mid-tier universities, had been cut by two-thirds. Accounting majors at programs outside the top 25 saw placement rates fall below 50%. The CPA pipeline crisis the AICPA had worried about for years was solved: the need for the pipeline was eliminated.

The MBA followed. Two-thirds of less selective programs had already reported declining applications in 2025³⁰, before AI became the primary narrative. Corporate middle management, the layer that MBA programs existed to feed, was being hollowed out, and the value proposition of a $180,000 MBA outside the top 20 simply ceased to exist.

There was nothing left to cross-subsidize with.

On March 25, 2026, the First Lady of the United States walked a red carpet in the East Room of the White House alongside a humanoid robot. The machine, produced by the California startup Figure AI and designated Figure 03, introduced itself to an audience of foreign dignitaries' spouses in eleven languages. Melania Trump then asked the room to imagine a humanoid educator named "Plato." Plato would be patient, always available, and able to deliver personalized instruction in literature, science, philosophy, and mathematics in the comfort of a family's home. Education Secretary Linda McMahon smiled from the front row.³¹ The American Federation of Teachers called it "every parent's nightmare." The administration called it the future.

By the fall of 2028, the individual crises (demographic, technological, financial, regulatory) had fused into a single self-reinforcing feedback loop. AI displaced white-collar workers. The entry-level positions that justified expensive degrees were evaporating. Students stopped enrolling. Universities lost revenue. They cut programs and staff, narrowing what they could offer, further undermining the case for attendance. To sustain enrollment they discounted deeper, which eroded net tuition revenue further. More closures followed. Each closure shook confidence in the system.

The loop had no natural brake.

By 2030, the hierarchy that had organized American economic life for half a century had partially inverted. A licensed electrician with a two-year certificate earned more, with greater job security, than a CS graduate from a non-elite university. A plumber was more insulated from disruption than a paralegal. A nurse practitioner had better career prospects than an MBA from a school ranked outside the top 20.

Community colleges experienced a genuine renaissance. As four-year degrees lost their signaling value, demand surged for credentials that led to work AI could not do. Enrollment at public two-year institutions rose roughly 12% between 2027 and 2030, even as four-year enrollment cratered.

By summer of 2030, the landscape had resolved into a brutal clarity. What survived told us what a university actually was, once the vocational scaffolding was stripped away.

The elite research universities survived and consolidated power. Harvard's $50 billion endowment could sustain the institution indefinitely with zero tuition revenue. Stanford, MIT, Princeton, and their peers were insulated by endowments, brand equity, and the self-reinforcing logic of elite credentialing: the more chaotic the economy, the more desperately families sought the safety of a name-brand degree. Acceptance rates, already below 4%, fell below 2%. These were less universities than luxury brands with research operations attached, and the crisis widened the moat. MIT restructured around AI-human collaboration. Stanford launched an AI Residency program. They had the resources to experiment and the brand to attract students regardless. The irony was thick: the institutions that needed to change the least were the ones most capable of changing. The ones that needed radical reinvention had no resources to attempt it.

Universities with genuine research capacity in fields essential to the AI buildout (materials science, power systems, semiconductor design, the mathematics of machine learning) found new funding models. Corporate money partially replaced declining federal support, though it came with strings that made faculty uncomfortable and raised legitimate questions about intellectual independence. The institutions that navigated this were ones with something the private sector couldn't easily replicate: deep expertise, long time horizons, willingness to pursue work that might not pay off for a decade. Georgia Tech, Purdue, and the University of Michigan were schools with strong engineering programs and established industry relationships. They found leverage they hadn't anticipated.

A handful of small institutions survived by being genuinely, irreplaceably distinctive. Hillsdale College, with its refusal of all federal funding since the 1980s and its unflinching commitment to the classical liberal arts. St. John's College, with its Great Books curriculum that had never pretended to be vocational in the first place. A few others. Their survival proved a thesis: that there is a market for education understood as formation rather than credentialing. But it simultaneously was proving how small that market is.

Community colleges and trade programs were the smartest bet in higher education. Low tuition, flexible scheduling, vocational credentials leading to work AI could not do.

Everything else was in trouble.

The rate of closures since 2028 has commanded the headlines. But the closures are not the story. The story is what is happening right now at the two thousand institutions between Harvard and Northland College: the large public flagships, the mid-tier privates, the regional comprehensives with endowments of $500 million to $5 billion and enrollments of 10,000 to 40,000. These schools are not closing. They are being restructured under financial duress, and the process is still in its early stages.

For most of the decade, restructuring was easier to describe than to execute. The barrier was the accreditation regime, designed as both quality assurance and gatekeeping. In normal times the accreditation moat was an asset. In a disruption it became a prison. The standards required faculty-to-student ratios, physical plant, library resources, and governance structures designed for the model AI was making obsolete. Any institution that tried to restructure around AI from the bottom up triggered accreditation review, and review threatened Title IV eligibility, and that threatened survival. So the schools did what the rules permitted. They bought the AI tools and layered them on top of unchanged cost structures, paying for both models at once. Institutions that might have reinvented themselves died inside the framework, maintaining their faculty ratios and governance documents all the way to the closing ceremony.

The schools that survived did not break through the regulatory lock. They went around it. The enabling mechanism is financial exigency. AAUP guidelines permit the termination of tenured faculty when an institution faces genuine financial crisis. For decades, boards treated the provision as unthinkable. Then a few invoked it and survived the legal challenges. Then more followed. By the 2030–31 academic year, exigency declarations were no longer news.

At most universities, 60 to 70 percent of the operating budget is personnel, and tenure renders a significant portion of that a fixed cost. You cannot restructure a cost base you cannot touch. Until you can.

What follows exigency is amputation. We have watched entire departments being consolidated or eliminated. Programs whose graduates no longer have a labor market to enter are cut. Programs connected to licensing requirements and physical human presence, nursing, allied health, skilled technical fields, are expanded. The institution retains its name, its campus, its Saturday football games. It loses much of its previous academic character.

Tuition appears to be the primary dividing line for how fast the restructuring hits. Large public flagships charging $12,000 to $18,000 for in-state students have so far been buffered by both price and political protection. The governors of Indiana and Ohio have publicly declared they will not allow their flagship state institutions to fail, and we expect similar commitments from other states as the pressure builds. But state budgets are themselves under severe strain as the broader labor market contraction erodes income and sales tax revenue, and the protection of flagships will likely come at a cost. Regional campuses are increasingly sacrificed to preserve flagship institutions, a pattern Penn State set in 2025 when it closed seven satellite campuses while renovating its stadium. Community colleges will continue to be funded because they are the retraining vehicles that workforce legislation requires.

But everything between those two poles (the mid-tier privates, the regional comprehensives, the small denominational colleges with endowments under $500 million) will continue to occupy a political no-man's-land: too expensive to bail out, too numerous to save individually, and too far from the retraining mission to benefit from workforce funding. Within this exposed middle, then, the picture is uneven. Even the strongest privates in this band, the schools charging $50,000 to $65,000 with substantial endowments and loyal alumni, are showing price sensitivity that brand equity can no longer absorb. Donor engagement has been a meaningful buffer: at several institutions, major gifts have sustained programs and athletic departments through the worst of the transition. But donor capital follows sentiment, and as the restructuring reshapes the institutions donors remember, we are watching those commitments closely.

Some institutions in the middle have found real, if limited, savings by turning AI inward: financial aid processing, enrollment management, compliance reporting, student services. Administrative headcount reductions of 20 to 30 percent, achieved without touching a single faculty line, are buying time.

But efficiency gains are deferring the harder question, and the crisis is already answering it: distinguishing formation-as-marketing-copy from formation-as-institutional-commitment. Every university mission statement in America claims to develop critical thinking, ethical reasoning, and the capacity for lifelong learning. For most, that language has been decorative, wrapped around a fundamentally vocational product. A handful of institutions have meant it, building robust core curricula, investing in mentorship, sustaining communities that were real and not merely brochure copy. But the crisis is stripping the decoration away. What remains will either be a genuine institutional commitment to the cultivation of persons, or nothing.

Our expectation is not that the vast middle will vanish. But what emerges from this decade will be smaller, leaner, vocationally narrower, and unrecognizable to anyone who attended these institutions a generation ago. Whether that constitutes survival depends entirely on what you think a university is for.

The university, at its origin, was not a job-training program. The medieval university was founded on the conviction that the pursuit of truth was itself a worthy human activity; not because it produced economic returns but because it was constitutive of human flourishing. For a few centuries, the American university managed to serve both masters: it cultivated minds and prepared workers. The current crisis is fundamentally severing those two functions. AI has taken the second. The question is whether the first can survive without it.

What remains, when the vocational scaffolding falls away, is something closer to what higher education was before the GI Bill transformed it into a mass industry: a place for the cultivation of persons. A place where young people encounter ideas that challenge assumptions, form communities that shape character, and confront questions like, "What is a good life? What do I owe others? What is my purpose?" These are questions that artificial intelligence cannot answer. They are not questions about information, but about formation.

That probably sounds like a luxury for the wealthy. The truth is, without massive reinvention, it is. The First Lady had already named the replacement. She called it Plato.

The pre-industrial university with faculty-to-student tutoring was a preserve of the elite. The post-AI university may revert to something similar. The great democratic experiment of mass higher education was founded on the idea that a rigorous education should be available to anyone with talent and desire. This may turn out to have been a historical anomaly, enabled by a specific set of economic conditions that AI is now dismantling.

It is possible to imagine a future in which AI-augmented education is radically cheaper and more accessible, in which the cost barriers fall along with the vocational justification. It is possible to imagine institutions that reinvent themselves not as credentialing factories but as communities of inquiry oriented toward human flourishing. AI 'faculty' could be used for content delivery while human faculty focus on mentorship, dialogue, and the friction of one mind against another.

But imagining that future and building it are different things. The accreditation system resists it. The financial models don't support it. The political will does not exist. And the institutions best positioned to attempt it are the ones that don't need to.

The $700 billion education system as it has existed since the mid-twentieth century was built on a bargain. Families paid. Institutions credentialed. Employers hired. Each party's participation depended on the others, and for three generations the cycle held. AI has broken the employer's side of that bargain, and without it, the other two cannot sustain themselves. The universities will not all close. But the bargain that made them a mass industry is over, and so far nothing has emerged to take their place.