- Binding constraint is the PhD pipeline, not grants
- Industry co-funding captures the training function
- By 2035: >60% of AI-adjacent PhD stipends at top-30 R1s flow from industry fellowships rather than federal training grants
- Universities become de facto CROs for Big Tech ("national lab without the charter")
- Second-order: mid-career faculty exits to industry labs ~double, hollowing mentorship — the one research input AI can't substitute (apprenticeship requires a human who has done the work)
- This explains why hub/teaching split is stable: hubs stay alive as long as apprenticeship is worth pricing; AI raises value of human mentorship in physical/lab science while destroying it in lecture
- Falsifier: if federal training grants still fund majority of STEM PhDs in 2035 AND faculty-to-industry exit rates flat vs 2025 baseline

Instructions:
- One Markdown note, single dated claim
- # title = claim in one line
- 1-3 short paragraphs
- End with "What would change this: …"
- Source or Unverified: for load-bearing numbers
- Don't mention tools, sessions, passes, or note-writing
- Don't invent evidence
- If only planning, output nothing — but there's substance, so write.

What can I source honestly?
- Federal funding share of university R&D: real fact — federally funded share of university R&D has declined from ~70% in the 1960s-70s to roughly 50-55% recently (NSF HERD survey). I can cite NSF Higher Education R&D (HERD) survey for the baseline that federal share of academic R&D has been drifting down while industry-funded share sits around 5-6%. That's real: HERD 2023 shows federal ~50%, industry ~5-6% of academic R&D expenditures.
- Industry PhD fellowship growth at AI-adjacent fields: there are real reports of industry funding rising in CS PhDs — the CRA Taulbee Survey shows that in computer science, industry funding of CS graduate students has grown; by early 2020s industry support for CS grads was substantial (Taulbee data showed ~38-55% of CS PhD students supported by industry sources in recent years? Actually Taulbee 2022-2023: industry funding for CS graduate students exceeded federal? Let me recall: CRA Taulbee Survey reported that in 2022, industry was the largest single source of support for CS PhD students — around 38% vs federal ~30%? I recall the 2021-2022 Taulbee showing industry support for graduate students surpassing federal support in CS. There was coverage (e.g., "industry now funds more CS PhD students than the federal government" — I believe CRA Taulbee 2022 showed industry ~38.6%, federal ~34%? Something like that. To be safe I can say "CRA's Taulbee Survey already shows industry support for CS PhD students rivaling or exceeding federal support" and tag as approximately real. Hmm, risk of inventing. Safer: "Unverified:" for the >60% forecast obviously — it's a forecast. For baseline, I can cite NSF HERD for federal share of academic R&D ~50% and industry ~5-6% (real). And Taulbee for CS-specific industry support rising — I'm fairly confident Taulbee shows industry as a major/f growing source for CS grads. I'll phrase carefully: "CRA's Taulbee Survey already shows industry support for CS PhD students rivaling or exceeding federal support" — I believe this is true for computer science. Actually the CRA Taulbee 2021 survey: "Industry support for graduate students (35%) exceeded federal (33%)" — I recall headlines like that around 2022. I'll keep it hedged with "roughly" or tag it. Better to hedge: "by the early 2020s, CRA's Taulbee Survey showed industry support for computer science graduate students rivaling federal support — Unverified as exact share." Hmm, that's awkward. I can name Taulbee and give approximate claim; the instruction says name a source for any load-bearing number OR start sentence with Unverified. I can do both: name Taulbee and mark the exact share Unverified.

The >60% by 2035 forecast: that's the claim itself — it's a dated forecast; mark as forecast/Unverified (it's inherently a prediction, doesn't need a source, but I should note it's a forecast).

Faculty exits doubling: forecast, Unverified.

Mentorship/apprenticeship claim: conceptual argument, no source needed