Endowment or DebtWhat is actually breaking in the college-to-job model, what is not, and the three functions that need new carriers

Two things are true at once, and almost everyone writing about this holds only one of them. The degree still pays. The path to using it has become unreliable. Those are not in tension — they describe different parts of the same machine, and only one of them is broken.

The sovereign arrived at both ends

Start with what happened rather than with what people are saying about it. In July 2025 a single piece of legislation did two things that are usually discussed separately, and the fact that they came from the same statute is the story.

It raised the endowment excise tax from a flat 1.4% to a tiered structure reaching 8% on the wealthiest institutions — Harvard, Yale, Princeton, Stanford and MIT all in the top tier, with Princeton's projected bill running from $39 million to roughly $223 million a year.

And it closed the federal borrowing channel underneath everyone else. Effective 1 July 2026, Grad PLUS ended for new borrowers and Parent PLUS was capped at $20,000 per student per year and $65,000 in aggregate, with existing borrowers grandfathered for three years or programme completion.

One statute. The endowed tier taxed, the tuition-debt tier's credit line closed. These are usually reported as separate stories — one about elite universities, one about families — and they are the same story. The sovereign arrived at both financing models in the same session.

Which means the pressure most commentary is describing — higher rates squeezing families — is real but lagged, arriving one cohort at a time through next May's auction. The pressure almost nobody is describing is already here: a family facing a $70,000 sticker price can no longer close the gap with federal credit, and must find private borrowing, more aid, or a different institution.

Rates are the weather. The caps are the climate.

What this does not establish: whether the caps reduce enrollment, shift it downmarket, or are simply absorbed by private lenders. That is not yet observable and this page will not guess at it. What is observable is that the channel narrowed, by statute, on a fixed date.

What is actually breaking

The entry rung. As of the second quarter of 2026, unemployment for recent college graduates stood at about 5.7% against 4.1% for the workforce overall — an inversion that is genuinely historically unusual. Forty-two percent of recent graduates work in jobs that do not require a degree.

And underemployment turns out to be sticky in a way that makes it a life outcome rather than a rough first year. Analysis of 10.8 million graduate career histories found 52% underemployed one year after graduating — and of those, 73% were still underemployed a decade later. The first job is not a starting point. It is, for most people who land badly, the shape of the next ten years.

The earnings consequence is where the split appears:

The premium is not shrinking so much as splitting. The average is still excellent and increasingly describes nobody.

The guard: what is not breaking

Three claims circulate constantly and are false. Anyone building an argument on them will lose it.

Enrollment is not collapsing. It is growing. Undergraduate enrollment rose to 16.2 million in autumn 2025, up 1.2%, the fourth consecutive term of growth. Community colleges rose 3.0%. Undergraduate certificates grew 10.2% and hit a decade high.

The degree has not stopped paying. The median graduate earns an annual return of about 12.5% on the investment, a figure that has held between 12 and 13 percent for three decades — better than the long-run return on equities. Bachelor's holders earn roughly 63% more per week than high-school graduates.

Employers did not abandon the degree. They rewrote job advertisements. Analysis of 316 million postings found roles that dropped a degree requirement hired about 3.5 percentage points more non-degreed workers — but only 3.6% of roles dropped one. The net effect: 0.14 percentage points, or in the researchers' words, "not even 1 in 700 hires." Nearly half the firms that announced the change were classified as adopters in name only. And the share of postings requiring a bachelor's has been rising again since early 2024.

And on AI as the cause — the honest answer is that nobody knows yet

The most-cited study finds employment for 22-to-25-year-olds in AI-exposed occupations running roughly 19% below where it would be had it tracked less-exposed peers. It is a striking finding and it should be treated carefully: it is an unreviewed working paper; its authors state plainly that it "does not estimate a causal impact"; the effect nearly vanishes in nationally representative survey data (a gap of −0.022 with a confidence interval spanning zero); and it roughly halves when education is controlled for. Several independent analyses find the hiring turn began months before ChatGPT existed, pointing at the March 2022 rate shock instead. A 2026 synthesis by economists including a former BLS Commissioner concluded there is "little evidence that AI is causing significant job losses right now."

The framework's reading is that AI is a real pressure on the entry rung. The framework's reading is not that this has been demonstrated, and this page will not pretend otherwise.

The cliff and the rise — both are true

As this page went up, the news cycle filled with a claim that sounds like it contradicts everything above: colleges are about to run out of students. It does not contradict it. It completes it — and it is worth being precise, because both halves are real.

Enrollment is rising now. Four consecutive terms, driven by adults over 25, community colleges, and short career-linked certificates — students hedging inside the system.

And the pipeline behind it is structurally shrinking. WICHE projects that US high school graduates peaked at roughly 3.8–3.9 million in 2025 and will fall about 13% through 2041 — roughly 576,000 fewer students. Beginning in fall 2026, the birth decline of the 2007–09 recession arrives on campus, eighteen years later, exactly on schedule. The decline is structural, not cyclical: those students were never born.

Colleges are not running out of students. They are running out of births — a regeneration failure invoiced eighteen years later.

Which is why this section belongs on this page rather than in the news. The fertility half of this framework was already measured: The Circulation, Article 6 traces total fertility from 2.12 to 1.60 and separates the two gaps that produced it — the fall in intended family size, and the widening gap between what people intend and what they realize. The demographic cliff is that finding arriving in higher-ed finance. An economy that stopped converting its accumulation into the next generation eventually finds every institution priced on the next generation's arrival — colleges first, because their customer is eighteen years old by definition.

It also closes the loop on the funding squeeze above. Small tuition-dependent colleges are closing while national enrollment rises because they sit exactly where the two curves cross: dependent on the shrinking first-time pipeline, without the endowment layer that absorbs a demand shock. Consolidation at the bottom, adults and certificates filling the top — that is what a system looks like when demand holds while its demographic base contracts.

The endowment test

Put the cliff, the AI commoditization of teaching, and the funding squeeze together, and the institutions sort into two species — and the sorting variable is the one this page is named for.

The endowed tier is insulated from the market. Princeton's endowment provides roughly two-thirds of its operating revenue; Harvard's provides 37% — a $2.5 billion annual draw; Yale distributed $2.1 billion last year. Their income does not depend on selling the thing AI made free, and the demographic cliff empties the marginal college's classroom, not the selective one's queue — an institution rejecting 96% of applicants absorbs a 13% smaller pool by reaching slightly deeper into a line that still wraps around the block.

The tuition-debt tier is not. The institutions closing — sixteen announced last year, small tuition-dependent privates first — are precisely the ones whose entire revenue model was a bet on the entry-level wage of an eighteen-year-old who is now, increasingly, not born. Not one closure headline names an endowed university.

And here the endowed tier's actual vulnerability comes into view, which is not economic. A position in a society rests on three separable supports — social acceptance, legal equality, and economic participation — and a position resting on fewer than all three is fragile in a way that stays invisible while conditions hold. The endowed university has the third in abundance and is losing the first. Economically successful, and targeted precisely because of that success is a recognisable historical position, and it is not one that wealth defends. And then the second shoe drops, and it is not a market force. In July 2025 the endowment excise tax rose from a flat 1.4% to a tiered structure reaching 8% on the wealthiest institutions — Harvard, Yale, Princeton, Stanford and MIT all in the top tier. Princeton's projected bill: $39 million to roughly $223 million a year. Stacked on the research-funding withdrawal, this is the sovereign arriving at the endowment layer.

The inversion underneath

Now the part that actually matters, and it is not about college at all. It is about who carries the risk when the rung breaks.

This is not a metaphor. It is documented in stone.

In eleventh-century South India, a Vedic college at Ennayiram supported 340 students and 14 teachers on the revenue of 45 veli of land — an annual requirement of 10,506 kalam of paddy, recorded with the daily grain allowance for each grade of student. At Tirumukkūḍal, an inscription of 1067 records a college of 60 students alongside a fifteen-bed hospital, itemising the salary of the physician, the surgeon, two male nurses, two female nurses, the barber and the waterman, along with sixteen named preparations and their quantities.

The mechanism is the point: land revenue alienated by rulers to temple assemblies, which then paid for teachers, students, physicians and staff on defined shares. Not fees. Not loans. The student's access did not depend on the student's own future earnings.

The question is not whether to go back. Nothing goes back. The question is what carries the functions once a single institution stops carrying all of them at once.

The fuller reading

This page uses the temple material for one narrow purpose — the financing structure — because that is the part that bears directly on student debt. And it is cited as a record, not a template. Ennayiram and Tirumukkūḍal are eleventh-century, which on this framework's own reckoning falls inside ascending Kali (499–1698 CE) — the lowest phase of the upward arc, not a higher age. The inscriptions matter precisely because they are late: they are the last clear documentation of an architecture that no longer had the conditions to sustain it, recorded before the capture completed. The claim is re-emergence under returning conditions, not restoration of an eleventh-century institution — and the financing mechanism, income from durable capital against a bet on an unearned wage, can be judged on its own arithmetic regardless of which century noticed it. The wider argument, which reads the temple as an integrated operating system running nine functions through one architecture, and traces its degradation through internal corruption, colonial capture and post-independence bureaucratic capture, is developed at length elsewhere: What the Temple Actually Was.

Its closing formulation is the one worth carrying, and it is the reason this page is not proposing a restoration: the temple is not the building — the temple is the integration. What is being asked for is not the institution back. It is carriers for the functions the institution held together and that nothing now holds together.

What the work in the temple actually was

There is a detail in the inscriptions that looks like a footnote and is not. The temple servants were paid in parcels of land. Historians read this as implying the temple work was part-time — and treat it as a qualification on the "temple as employer" claim.

Read it the other way and it stops being a qualification. The land was the livelihood. The temple work was the demonstration.

A craftsman, a musician, a physician, an accountant did their work at the temple as offering — unpaid or barely paid, to the highest standard they were capable of, in the one place where the entire region would see it. And from there they were engaged. The temple was not only where the work was given. It was where the work was seen.

What makes this more than a nice observation is that it names something a degree never did and that nothing currently does: a shared venue with a known standard. Bronzes were cast to canonical proportions set out in the Shilpashastra. Everyone judging the work was judging against the same measure, in the same place, over years. That is a very high-quality signal, and it is cheap to produce and impossible to fake.

Why this matters more now, not less

Because the artifact has stopped being evidence. In a ten-year study of nearly five thousand undergraduate reports, AI's signature appeared clearly after 2022 and not one student disclosed using it. You can no longer infer capability from a finished piece of work — which is precisely what a résumé, a portfolio, and a graded assignment all are.

The modern version of the mechanism partly exists and works imperfectly. Studying roughly 22,900 developers, researchers found that career concerns measurably increase open-source contribution during job search, shifting toward visible projects and high-value skills; a GitHub survey found about 70% of employers view such contributions favourably.

The signal was never the certificate. It was the work, done in the open, over time, in front of people who knew what good looked like.

The bundle was never one thing

There is a second argument running alongside the financing one, and it arrives at the same structure from inside the institution rather than from the ledger. Ian Bogost, writing in The Atlantic in September 2026 from a chair at Washington University, puts it this way: “Higher education is a bundle of unrelated things held together by custom, history, and accident.”

His bundle is credentialing, coming-of-age, and scientific research. Three crises pulled them apart: COVID proved credentialing works with campus life reduced or absent; AI severed the credential from the process, so that “social and academic progress are no longer woven together”; and the research-funding cuts exposed that education and research were never one enterprise.

The single best piece of evidence in that argument is one word:

Why this page cites a critic rather than rebutting one. Bogost reaches unbundling from inside a research university. This page reached it from the financing structure. The federal loan tables reach it a third way, by forcing the question of what exactly the borrowing purchases. Three independent routes to the same finding is worth more than any of them alone — and it is the kind of corroboration that survives a reader who dislikes the conclusion.

And he supplies a bundle member this page missed. The three functions below — formation, orientation, access to productive assets — contain no equivalent of the research mission, which is the largest single flow of money through the sector and the one least connected to undergraduates. That is a real gap in the analysis, named here rather than quietly patched.

Where the two arguments differ, and it matters: Bogost does not predict collapse. He writes that colleges will continue, and says plainly elsewhere that elite schools, large state schools and elite liberal-arts colleges will be fine. His claim is that “the appearance of college can endure, even as its spirit — a community organized for the education and formation of young people — comes undone.” That is close to this page's reading, and anyone citing his title as an obituary has not read past it.

Three functions, looking for carriers

The college-to-job model bundled three things that had never previously been bundled, and did it so successfully that most people stopped noticing they were separate.

Formation — becoming a person capable of holding complexity, before any question of employability. Orientation — knowing what to do next, especially when the path you were on changes underneath you mid-life. Access to productive assets — a claim on something that produces, rather than a wage alone.

The pipeline is failing at the third and straining at the first. And unbundling is not a prediction here — it is already visible in the underemployment persistence figure. A degree that no longer reliably delivers the asset still delivers the formation, which is exactly why the premium holds while the path frays.

What follows is what we are building against each function. It is offered as work in progress, not as a solution.

Education was a service before it was an industry

Which leaves the question underneath every section above: what is actually ending, and what is actually arriving? The framework's answer — held as reading, supported at every joint by evidence already on this page — is that the education industry is in its long decline, and education as service is slowly re-emerging. Not an event. A succession, of the kind every organizing framework before it has undergone: the model solved its era's problem, absolutized its solution, and is now being succeeded by what its own success made necessary.

What the industry sells is information transfer — and the price of information transfer has been falling toward zero for twenty years. Khan Academy is free. MIT's OpenCourseWare is free. Wikipedia is free. AI finishes the sequence: the marginal cost of a competent explanation is now nothing. An industry whose revenue rests on charging for what costs nothing to reproduce does not collapse overnight — it declines the way every such industry has declined, defending its gate while the gate loses meaning.

What cannot go to zero is formation — and here is the historical fact this page has been circling: the tradition never priced it. Ennayiram charged no fees; the endowment carried the teachers. The guru-dakshina came at the end, voluntary, after the giving — gratitude, not a price. Teaching was the dharma of the learned, not their business model. The fee-for-credential structure is roughly a two-century industrial-era artifact, built when education was retooled to produce standardized labor — and it is exactly as old as the wage-bet financing that is now breaking.

The motivation evidence suggests the artifact was costly all along: across 128 experiments, expected contingent payment undermines intrinsic motivation for already-meaningful work — and teaching is the canonical intrinsically-motivated profession. The verse and the meta-analysis agree. The offering-stance produces the better teaching.

And the succession is already visible at the edges, which is where successions always start. Free course platforms carrying the content layer. Homeschool networks carrying formation on family and community time. ESA and TEFA programs — whatever one thinks of their politics, and carrying the opposite fragility to the endowment's: socially accepted in the communities that use them and legally dependent on a programme a legislature can revoke, which is acceptance without guarantee — functioning structurally as public endowment disbursed per child: formation money routed to families rather than invoiced against a student's future wage. Even this operation's own model runs on the returning structure: every JyoLing course free, forever, with payment only for the living tutorial relationship — and students earning back by teaching others. The knowledge is given. What is priced is the relationship, which is the one thing that was always honestly priceable.

The industry charged for the knowledge and threw in the formation. The successor gives the knowledge and is financed for the formation. AI did not invent that model. It made every other model unpriceable.

What this argument does not claim

It does not claim colleges die now. The claim is succession, not funeral: the industry model is in a long final decline measured in decades, and the service model emerges slowly at the edges — free platforms, homeschool networks, award programs, endowed institutions. Both will coexist for a generation. Institutions on the wrong financing model will feel it first; most students alive today will still be taught inside the declining structure, and it will still work for many of them.

It does not claim college is over. The premium is near a three-decade high, enrollment is growing, and 74% of parents still want a postsecondary pathway for their child. Anyone who tells a young person otherwise is doing them harm.

It does not claim these products are proven. JyoLing Academy has not been through a randomized trial. Neither has anything else in this space that is being sold to you with confidence.

It does not claim the old order should return. The endowment structure is worth understanding. The caste order that financed it is not a model, and reading it as harmonious is a documented historical error with real consequences.

And it does not claim AI caused this. The evidence is contested and the timing is awkward for the simple story. What is certain is the shape of the break, not its cause.

What the framework does hold is narrower and, on the evidence above, defensible: a system that finances formation by borrowing against a wage that has not yet been earned is structurally fragile at exactly the moment the wage becomes unreliable. That moment has arrived for a large enough minority of graduates to be visible in national data. The response is not a better credential. It is carriers for the functions the credential was standing in for.

Common questions

Is the college-to-job model ending?

No — the entry rung is breaking while the degree still pays. Recent-graduate unemployment reached 5.7% against 4.1% for the workforce, and 42% of recent graduates work in jobs not requiring a degree. But the median graduate still earns a 12.5% annual return, near a three-decade high, and enrollment has grown four consecutive terms.

Is college still worth it?

On average yes, but the average conceals a split. A graduate in a college-level job earns about 88% more than a high-school graduate; an underemployed one earns about 25% more. At the 25th percentile the return falls to 2.6%, and about 23% of bachelor's programmes show a negative return once non-completion risk is counted.

Are graduates worse off than people without degrees?

No. Against workers of the same age without a degree, graduates sit at 5.7% unemployment against 7.2%. The comparison that produces the alarming version includes older workers with decades of tenure.

How was education funded before student loans?

By durable endowment. In medieval South India, land revenue alienated to temple assemblies paid for teachers, students, physicians and staff on defined shares — a college of 340 students at Ennayiram, a college and fifteen-bed hospital at Tirumukkūḍal with every salary itemised. Access did not depend on the student's own future earnings. It depended on the endowment, and on being born into the caste it served.

What replaces the pipeline?

Not one institution but carriers for three functions it bundled: formation of a person before employability, orientation for adults whose path changed mid-life, and access to productive assets rather than a wage alone. Nothing currently bundles them the way a degree did.