Student Loans for Degrees AI Can Replicate: The Economic Model Breaks

There is a graph you should look at. It plots two lines from 2010 to 2026. The first line tracks the average annual tuition at four-year private universities in the United States, adjusted for inflation: it rises steadily from $28,000 to $41,000, a 46 percent real increase over sixteen years. The second line tracks entry-level salary premiums for college graduates relative to high school graduates in the same fields: it rises from 2010 to 2018, peaks, and then declines in the fields most affected by AI capability — writing, analysis, data work, basic legal and accounting functions. The lines don’t cross yet. But they are converging.

This convergence is the economic story underneath the AI-in-education debate. Most of that debate focuses on pedagogy: how should teaching change, how should assessment change, how should institutions adapt. These are important questions. But the financing question is more urgent, because it involves $1.9 trillion in outstanding debt, 45 million borrowers, and an underlying asset — the credential — whose value is falling in ways that no one in the lending apparatus has officially acknowledged.

The federal student loan program was built on a specific empirical claim: that a college degree produces a wage premium large enough to justify the cost of acquiring it. This was true, on average, for most of the program’s history. The premium was large enough to offset even substantial debt for most borrowers in most fields. The program’s defenders point to this history. What they don’t address is that the premium is empirical, not structural — it exists because employers use the credential as a sorting signal, and that sorting function is being disrupted.

Sorting signals work when they are costly to fake. A college degree was costly to fake in 1995. You had to show up for four years, pass the courses, and have the social and financial resources to sustain the effort. Employers could not easily verify underlying competence, so they used the credential as a proxy. The proxy worked because the correlation between credential acquisition and genuine competence was high enough to be useful.

AI doesn’t fake the credential. It fakes the underlying competence. A person with a bachelor’s degree in marketing who has used AI to pass their courses arrives in the job market with a credential that employers still trust. They may lack the underlying competence the credential is supposed to certify. The employer discovers this gradually, after hiring. The credential sorting mechanism is producing noisier signals, and the noise is concentrated in exactly the fields where AI assistance with coursework is easiest.

The lenders know this at some level, which is why income-driven repayment enrollment has increased 34 percent since 2024. IDR was designed as a safety net for borrowers who encounter hard times. It is becoming the primary repayment mechanism for an entire generation of graduates in fields where the promised wage premium has not materialized. When the safety net becomes the standard path, it’s a sign that the asset being financed isn’t delivering its promised return for a significant fraction of borrowers.

The for-profit college sector deserves its own moment here, because the story of for-profit colleges and AI is both faster and uglier than the story of nonprofit universities. For-profit institutions were already under regulatory and reputational pressure before AI became a factor — the Obama-era gainful employment rule, the mass closures of ITT Tech and Corinthian, the settlements over deceptive marketing. The sector had been contracting since 2016.

AI gave it a brief, perverse resurgence. Between 2024 and 2025, several for-profit operators launched programs built explicitly around AI — “AI-assisted learning,” “AI-powered bootcamps,” accelerated credentials delivered primarily through AI tutoring systems. The pitch was: get a credential faster, cheaper, with our AI system guiding you. Enrollment grew. Revenue grew. The underlying quality of the credential being delivered was, in many cases, even lower than before, because the AI was doing work the student should have been doing.

The Department of Education is now dealing with a new wave of gainful employment investigations targeting programs whose graduates’ income outcomes don’t justify the debt. This was entirely predictable. It was predicted. Nobody with regulatory authority did anything fast enough. The students who enrolled in 2025 are discovering in 2027 that the credential they paid for and borrowed for doesn’t get them the jobs they were shown in the marketing materials. The AI made the pitch more convincing. It did not improve the underlying program.

The community college sector is having a complicated moment that gets insufficient attention. Two-year programs, certificate programs, and workforce training programs at community colleges are seeing enrollment growth — that 11 percent figure from the previous discussion. This makes sense for several reasons. Community colleges are dramatically cheaper. Their programs are more directly vocational. The credential-to-employment pipeline is shorter and more transparent.

But community colleges are also the institutions with the fewest resources to invest in AI integration. A community college in rural Indiana with 2,400 students and a $22 million annual budget is not hiring AI curriculum consultants. Its faculty are teaching four or five courses a semester, often with heavy adjunct reliance, and don’t have time to redesign their courses around AI even if they knew how. The institutions that could most benefit from AI efficiency gains are the least positioned to capture them.

There is a policy intervention that would make a significant difference here: fund AI adoption specifically at community colleges with the same energy that the Department of Education funds broadband access and Title IV compliance. The money exists in various discretionary programs. The will to direct it toward the institutions that serve the most economically vulnerable students does not appear to exist, at least not in any budget I’ve reviewed.

The apprenticeship model is having a moment that should be bigger than it is. Apprenticeship programs in skilled trades, healthcare support, and technology have been growing — the Department of Labor reported 780,000 new apprenticeship registrations in 2026, a 23 percent increase over 2023. Apprenticeships pair learning with earning. The student doesn’t take on debt to acquire a credential. They earn while developing skills that employers have directly validated through the program design. The credential at the end is not from an institution that may or may not have taught them anything. It is from the practice of doing the work.

Apprenticeships also happen to produce skills that are AI-resistant in different ways than university degrees. The HVAC technician, the medical sonographer, the industrial electrician — these roles require physical presence, manual skill, and real-time judgment in environments that vary constantly. AI can assist with diagnostic information and documentation. It cannot wire the panel or calibrate the equipment. The credential for these roles reflects competence that is genuinely difficult to fake, because the competence is demonstrated in practice.

Germany figured this out decades ago. The dual education system — school plus apprenticeship, deeply integrated — produces skilled workers without the debt burden that defines American post-secondary education. The vocational track in Germany is not a consolation prize for students who couldn’t get into university. It is a respected, well-compensated pathway. The cultural shift required to establish that in the United States is enormous. But the economic logic points in that direction increasingly clearly.

What would it take to fix the student loan financing crisis before it produces a genuine economic shock? The honest answer is: nothing that any constituency with political power wants to do. Loan forgiveness addresses the existing debt but not the structural problem. Interest rate reform reduces the ongoing burden but not the underlying asset risk. Free college proposals shift the cost to taxpayers without changing the fact that many of the degrees being purchased aren’t delivering their promised returns. The structural fix — matching credential costs to credential outcomes, refusing to finance programs that don’t produce wage premiums sufficient to repay the debt — requires institutional accountability that for-profit operators and many nonprofit universities would fight fiercely.

The market may do what policy won’t. If enough students opt out of expensive degrees, if the enrollment decline continues, if employers develop alternative signals that work better than credentials, the institutions that can’t justify their cost will lose enrollment and eventually close. This process is already happening at small private colleges — 42 closed in 2025 and 2026 combined, the highest two-year total since tracking began. It will continue. The question is whether it happens slowly enough to hurt a generation of students who borrow based on the old theory of credential value, or fast enough to force reform.

The students making enrollment decisions right now, in the spring of 2027, are making them with information that is incomplete and with advisors — high school counselors, parents, college financial aid offices — who have strong incentives to present the most optimistic interpretation of the data. The honest version of the college value conversation is not one that college admissions offices are having. The market for candid information about credential ROI in an AI economy is badly underserved. Someone should fill it. That it hasn’t been filled yet is its own kind of institutional failure.