NextFin News - Three stories dominated the latest Wall Street Week, and together they sketch a single, uncomfortable picture: the U.S. economy is being pulled in three different structural directions at once. National Economic Council Director Kevin Hassett argues that AI's productivity boom is running far ahead of what official statistics capture; America is spending billions to rebuild a rare-earth supply chain after decades of Chinese dominance; and universities are bracing for a "demographic cliff" as the college-age population shrinks. The common thread is not optimism or pessimism — it is a bet that the American economy can rewire itself faster than the data, the supply chains, and the students can keep up.
The Productivity Boom That the Data Can't See
Kevin Hassett, director of the National Economic Council, laid out the administration's case on Sept. 28 at an Economic Club of New York luncheon: absent external shocks, annual GDP growth will likely hit 4%, with the AI-driven productivity boom as the biggest contributor. "The studies that are out right now, academic studies, show that the labor market effect of AI is that firms that start using AI see their sales go up a lot, their employment go up a lot and wages go up a lot because the people are more productive," Hassett said. He has also argued that AI is creating jobs rather than killing them, even as most U.S. adults say they fear AI adoption will lead to job loss.
The claim is provocative because the official data do not yet show it. Measured labor productivity has been running around 2.5%, but utilization-adjusted total factor productivity — the Federal Reserve Bank of San Francisco's measure that strips out the effect of simply working existing workers and equipment harder — grew just 0.07% over the four quarters ending in the first quarter of 2026. In other words, almost none of the productivity gain so far is coming from doing things better; it is coming from working what we already have harder.
Hassett's argument, and a growing chorus of economists, is that the statistics are simply missing the AI economy. A research note from Epoch AI published Aug. 24 estimated that U.S. GDP growth is understated by about 0.3 percentage points because national accounts fail to capture most of the value created by fabless chipmakers, primarily Nvidia — and that the gap could widen to 2 percentage points by 2028. The mechanism is an accounting one: Nvidia designs chips in the U.S. but ships manufacturing offshore, so much of the value added leaks out of domestic production statistics even as the profits accrue to American shareholders. The Congressional Budget Office, in February 2026, for the first time added an AI productivity adjustment to its baseline — effectively betting on the optimists' reading to offset drags from trade frictions and immigration restrictions.
The measurement problem is not new. The statistical discrepancy between GDP and gross domestic income has been running wide for years — it hit 3.5% of GDP in the first quarter of 2022, the largest on record at the time — and in the first quarter of 2026 GDI ran 0.9% below GDP. That seam is where the missing AI value could be hiding.
There is a parallel argument inside higher education itself. University leaders contend that AI is reshaping how students learn and teachers teach, and that the ethical questions raised by the technology could create new opportunities for liberal arts education — precisely the fields that have faced the steepest enrollment pressure. If AI becomes a ubiquitous tool, the premium may shift from technical execution to judgment, communication, and ethical reasoning. That is an optimistic read, and it collides with the enrollment numbers: the institutions best positioned to sell a liberal arts premium are the elite ones that the demographic cliff barely touches.
The second-order question is whether the market is pricing a boom or a measurement error. If Hassett is right, the neutral interest rate is higher than the Fed thinks, because a more productive economy can absorb higher real rates without slowing. If the San Francisco Fed is right, the productivity narrative is a valuation crutch — and the companies priced for perpetual AI-driven margin expansion are vulnerable to a reality check.
"The studies that are out right now, academic studies, show that the labor market effect of AI is that firms that start using AI see their sales go up a lot, their employment go up a lot and wages go up a lot because the people are more productive."
The Rare-Earth Race: Decoupling on a Decade-Long Clock
While Washington debates whether AI is transforming the economy, it is also spending billions to transform the physical supply chain that makes AI — and defense systems — possible. Rare earth elements are the clearest case of a choke point: China accounted for around 60% of global rare-earth mining output in 2024, and roughly 91% of the separation and refining stages, according to the International Energy Agency. Malaysia is a distant second.
The U.S. has moved. Investment in rare-earth refining by the U.S. and Malaysia helped reduce China's share of that segment from over 90% in 2023 to 85% in 2025. But the IEA's Global Critical Minerals Outlook 2026, published in July, delivers a sobering number: even if every planned rare-earth refining project worldwide comes online on schedule, China's refining market share is only projected to fall to 70%-73% by 2035. This is not a sprint; it is a decade-long grind with the finish line still deep inside Chinese territory.
The vulnerability is not theoretical. China suspended some of its expanded export controls until November 2026, but the pause is a negotiating lever, not a settlement. Beijing already leveraged rare-earth export restrictions against Japan in 2026 over Taiwan. And the concentration runs deeper than rare earths: China processed 70% to 95% of global lithium, cobalt, phosphate, manganese, and graphite in 2025, and produced 98% of LFP cathode materials and 80% of global battery cells.
On the American side, MP Materials operates Mountain Pass in California — which the company says is now the world's second-largest rare-earth mine — with an operating refinery scaling up and a downstream magnetics business. But the U.S. still imports about 71% of its rare earths from China. The asymmetry is stark: America is betting its AI and defense future on a supply chain that will remain majority-Chinese for at least the rest of the decade.
The economics of decoupling are punishing. Building refining capacity outside China requires feedstock, cheap power, and a workforce trained in separation chemistry — all of which migrated abroad over decades. The U.S. and its partners are using government capital to bridge the cost gap, but the IEA's 2035 projection implies that policy can shave concentration, not eliminate it. The realistic endpoint is not independence but managed dependence: enough non-Chinese capacity to deter a total cutoff, but not enough to make one painless.
The Enrollment Cliff: A Cliff That Looks More Like a Hill
The third story is demographic, and it is already here. The phrase "enrollment cliff" dates to a 2012 report from the Western Interstate Commission for Higher Education (WICHE), and the fall of 2026 is when it starts to bite. WICHE projects that the national population of high school graduates peaked in 2025 at roughly 3.9 million and will decline 13% through 2041 — about 576,000 fewer students over the projection window. Nathan Grawe of Carleton College projects a 12-percentage-point drop in the number of 18-year-olds entering college from 2025 to 2030.
The cause is arithmetic, not preference: U.S. live births fell from more than 4 million in 2000 to 3.6 million in 2024, and the fertility rate dropped to 1.62, well below the 2.1 needed for population stability. As one observer put it, if they weren't born, they can't go to college.
But the impact is not evenly distributed. Overall enrollment actually rose about 1% in fall 2025 to 19.4 million — the third straight year of increases, though still below the 21 million peak of 2010. Community colleges gained 3%, public four-year institutions gained 1.4%, while private four-year institutions fell 1.6% and for-profits fell 2%. WICHE's projections show community colleges facing a 14-percentage-point decline in 18-year-olds in their entering classes from 2025 to 2030, regional universities 11 points, and national universities ranked 51-100 about 9 points. Elite institutions, ranked 1-50, face a much smaller hit.
There is a twist. The Common Application reported that as of Jan. 1, 2026, fall 2026 applications were up 4% from a year earlier, with the strongest growth among low-income, Black, and rural students. International applications, however, were down 7%. So the cliff is not a uniform collapse — it is a redistribution, and the institutions without brand or price power are the ones falling.
Geography sharpens the divide. WICHE projects that states in the Northeast, Midwest, and West will see double-digit percentage decreases in high school graduates through 2041, while the South is projected to gain 3% over the same period. Only 10 states are expected to grow from the 2025 peak. A regional university in the Rust Belt and a flagship in the Sun Belt are not facing the same cliff — one is staring at a fiscal emergency, the other at a manageable headwind.
The Synthesis: Three Bets on Rewiring
These three stories are not independent. They are three bets on whether the American economy can rewire itself — in how it measures output, in what it physically depends on, and in who shows up to do the work.
The cyclical-versus-structural call matters for each. The productivity debate is partly cyclical — measured productivity can revert as capacity utilization normalizes — but the measurement gap is structural: national accounts were built for a manufacturing economy, not a fabless-chip, intangibles-heavy one. The rare-earth decoupling is structural: once a refining ecosystem migrates, it does not migrate back on its own, and the IEA's 2035 numbers confirm that. The enrollment cliff is the most structural of all: the students who would have entered college in 2030 were already born, and no amount of marketing will create them.
The second-order implication is the one the market is not fully pricing. If AI productivity is real but mismeasured, and if the rare-earth supply chain stays concentrated, then the AI boom itself becomes a concentration risk: America's most important growth engine depends on inputs controlled by its principal strategic competitor. And if the enrollment cliff hollows out the regional universities that train the technicians and engineers, the labor force for both the AI buildout and the minerals rebuild shrinks just as demand for it peaks. The irony is direct: the administration betting on an AI-driven productivity boom is also presiding over a university system that may produce fewer of the very workers the boom requires.
The counter-thesis is straightforward and has powerful backers. The San Francisco Fed's utilization-adjusted TFP number — 0.07% — says the productivity boom is not happening at all, and that capital spending is simply inflating asset values. CSIS warned in July 2026 that betting government revenue projections on AI productivity is a "misreading of the data" that could produce misguided policy. On rare earths, the counter is that diversification is happening faster than the IEA's conservative scenario assumes, and that substitution — magnets that use less dysprosium, recycling, alternative chemistries — will ease the choke point. On enrollment, the counter is that Common App's 4% application increase shows demand is resilient, and that institutions can recruit older students, international students, and online learners to fill the gap.
The falsifying signals for the structural view are specific: if utilization-adjusted total factor productivity prints above 1% annually for four consecutive quarters, the "boom is imaginary" thesis fails. If China's share of rare-earth refining falls below 70% before 2030, the decoupling is ahead of schedule. If fall 2027 total enrollment holds at or above 19.4 million despite the smaller 18-year-old cohort, the cliff is shallower than projected.
What to Watch
Short term, watch the next GDP release for the statistical discrepancy between GDP and gross domestic income — it is the seam where the missing AI value is hiding. Medium term, watch whether China reinstates the full suite of rare-earth export controls after the November 2026 suspension expires. Long term, watch the fall 2027 enrollment numbers: they will separate the institutions with pricing power from the ones that thought they were elite.
The bottom line: America is betting on a productivity boom the data can't yet prove, a supply chain that will stay mostly Chinese for a decade, and a university system facing a demographic squeeze that no amount of AI can reverse. The winners will be the institutions, companies, and investors who recognize that all three rewirings are happening at once — and that they pull in different directions.
更多独家洞察尽在 nextfin.ai.
