# Tuition relief comes only from the teaching stack — the 25–30% of spend that AI can compress — so prices bifurcate: premium schools hold price, tuition-dependent schools cut toward cost

The cost-structure claim: AI attacks the smallest discretionary layer of university spend, not the largest. Instruction is roughly 25–30% of education-and-general expenses at four-year institutions (Unverified on exact share; the directional figure tracks IPEDS finance categories), and only the teaching portion of it — lecture prep, grading, TA staffing, adjunct sections — is substitutable. Research labs, administration, facilities, athletics, and accreditation compliance are untouched. So the total addressable savings is a minority slice of the cost base, and it arrives only as teaching roles turn over: tenured lines are sticky, so the compression phases in over a decade through non-renewal and non-tenure-track hiring rather than layoffs.

Consequence: tuition falls only where institutions must compete on price — tuition-dependent regional publics and small privates with no brand pricing power. Selective and flagship institutions capture AI savings as margin and reinvestment, not price cuts; net price at the top holds or rises through the demographic squeeze because their demand is inelastic. The result is a bimodal tuition distribution by the mid-2030s: a premium tier whose price signals scarcity, and a commodity tier where AI-assisted instruction at scale drives net tuition toward the cost of delivery plus a thin margin. The middle — schools with moderate brand, high tuition dependence, and full-price generalist programs — gets squeezed from both sides and is where closures concentrate.

The mechanism that dates it: savings require replacing teaching labor at hire, not firing it, so the effect is invisible in 2025–2029 (hiring freezes already slow), visible 2030–2037 as the demographic cliff forces tuition-dependent schools to actually price against AI-delivered alternatives, and locked in by 2040 when a cohort of institutions has built entirely around compressed teaching costs. This compounds the credential-side pressures already on record — cost relief lowers the price floor for the survivors even as demand for the generalist degree erodes beneath them.

What would change this: if instruction's share of spend is materially higher than ~30% at the tuition-dependent tier, or if accreditors force human-instruction minimums that cap substitution, the commodity tier never reaches cost-plus-margin pricing and tuition compression stalls — the middle survives on regulated pricing rather than dying on competitive pricing.