Let me compile the full durable inventory from all the notes:

**Core answer:**
- "Both" — the system bifurcates and thins; not uniform growth or shrink
- Total enrollment drifts ~19M → ~14–15M by 2045 (Unverified)
- Institution count falls while degrees conferred stay roughly steady — "thins, doesn't shrink"
- Median campus roughly halves
- AI is an accelerant/sorter of existing trends, not the primary cause
- Two-front squeeze: fewer 18-year-olds AND weakened degree signal
- Barbell: system shrinks at the median while top decile grows
- AI cost efficiency doesn't shrink the sector; it reallocates capacity — same efficiency monetized as headcount shrinkage in declining regions vs. growth capacity in growth states

**Demography (sourced):**
- Births 4.32M (2007) → ~3.59M (2023), ~17% decline (CDC/NCHS)
- HS grads peak ~3.9M/yr ~2025 (WICHE "Knocking at the College Door")
- Birth dearth hits college age 2029–2037
- NE/Midwest/Plains lose 10–20% college-age cohorts (WICHE); TX/FL/TN/GA/Carolinas grow
- WICHE ~13% decline in HS grads by 2041
- Net domestic migration to Sun Belt (Census Bureau)
- Regional divergence locks in demographically by late 2030s; AI converts headwind into closure rather than stagnation

**Regional/tier structure:**
- SEC-corridor applied flagships grow 30–50% headcount by ~2045
- Sun Belt gains across tiers (+10–25% seats through ~2040)
- Nonselective Northeast/Midwest contracts roughly one-third; closures concentrated PA, OH, IL, New England
- Tier-moderated not geography-absolute: Ivies/elite Northeast privates insulated
- The real dividing line: endowment depth (privates) + state-subsidy commitment (publics); region sets the demand curve
- Intra-state bifurcation: within growth states, money flows to flagships; tuition-dependent schools in SEC states can hollow out faster than Northeast peers
- By 2045 regional labels matter less than institutional balance sheets; the disruptable university is any tuition-first institution with AI-substitutable product
- Casualties concentrate in "sufficiency band": regional-public branches, SUNY comprehensives, tuition-dependent small privates below ~60% occupancy
- Program mix × region: licensed vs unlicensed share of credit hours = dominant predictor; licensed expansion takes 7–10 years (longer than declining-region runway); "program mix picks the winners, regional demography picks which losers can still switch"

**Credential value / signaling:**
- Screening vs insurance decomposition: screening dies first; insurance (blame-shifting/liability) persists through 2040s
- Degree survives as trust/insurance credential, not competence credential — AI can assay competence but not diligence/trust
- Selection-based signals are AI-proof (sorting at 18); certification-based signals are not
- Signaling migrates upward to selectivity/brand: "which degree" becomes the signal when "has a degree" stops discriminating
- Elite degrees gain value BECAUSE verification makes mid-tier signals redundant
- Licensed/statute-gated fields hold (nursing, law, accounting, teaching, engineering, medicine); licensure law pins ~20% of US employment to degrees
- Erosion hits by major before by institution; fastest collapse in generalist white-collar fields
- Entry-level/new-grad premium flattens first (youngest edge) — the scissors: new-grad starting premium and posting degree-requirements turn down while adult premium holds
- Non-selective institutions sell labor-market access → first-order elastic; selective bundle network/sorting → demand rises
- Erosion on slow clock, not cliff: ATS/compliance stickiness, negligent-hiring liability, generational turnover in hiring managers
- Degree reprices from entry ticket to exit credential / trust warranty by ~2035
- Residual degree value set by statutory gatekeeping, not brand signal
- Pre-AI drift evidence: Burning Glass "The Emerging Degree Reset" (~46% of middle-skill postings dropped degree requirements 2021–22); HBS "Dismissed by Degrees"; Stanford Digital Economy Lab "Can