The AI Consumer Shift

—The Big Insight

The most-watched AI investment trade is about who builds the technology: chipmakers, cloud platforms, and data centers. A second, larger trade is about who is still able to buy things once that technology changes who gets hired.

Three things make this more than a thought experiment. First, it is already visible in 2026 hiring data, not just in a forecast. Second, it compounds through a chain of second- and third-order effects — professional-services economics, political risk, retirement savings, education, and household formation — that most versions of this thesis stop short of. Third, it produces a testable, repeatable way to screen public companies, not just a narrative.

This report builds the case in that order: what is happening, why it compounds, what it does to spending, where that creates winners and losers, and what would have to be true for the thesis to be wrong.

Executive Summary

Source: Consumer Expenditure Survey, Current Population Survey, BLS occupational employment data, and the labor-market scenario modeled in Section 3.

01This Is Already Happening

Why it matters

A scenario is more useful when it is checkable against real data before it fully plays out. The entry-level hiring decline this report models is not speculative — it is already visible in hiring platforms' 2026 data, which both strengthens the case for paying attention now and disciplines how aggressive the assumptions should be.

Source: LinkedIn Grad's Guide 2026; Indeed Hiring Lab; Resume.org 2026 employer survey; Fortune / Federal Reserve Bank of New York new-entrant unemployment data.

Entry-level hiring fell roughly 6% year over year between December 2025 and February 2026, and entry-level postings on Indeed fell about 7% across 2025. In a 2026 survey, 21% of employers said they had already frozen entry-level hiring specifically because of AI, and 47% expected to eliminate entry-level roles entirely by 2027. Unemployment among new labor-market entrants peaked near 13.3% in mid-2025, the highest level in 37 years, before settling closer to 10–11%.

None of this proves AI is the only cause — that question is addressed directly in Section 11, because it changes how the thesis should be modeled, not whether it matters.

02The Mechanism

Most discussions of AI and jobs ask whether a whole occupation can be automated. That is the wrong resolution. A company does not need to dismiss its accountants, programmers, or analysts to transform the labor market — it only needs to hire fewer new ones each year while retaining experienced staff. The effect is invisible within any single company in any single year, but it compounds across a decade of graduating classes.

03The Labor Model: 8.3 Million Missing Careers by 2036

Setting up the chart

The model does not assume a sudden collapse in professional hiring. It assumes career-track placement rates decline steadily over roughly a decade as AI capability improves and employers redesign organizations around smaller junior teams. The gap between graduate supply and career-track demand is the "missing careers" figure.

What happens to the 8.3 million

Most displaced graduates still work. The defining feature of the scenario is not a 20% unemployment rate among all graduates — it is that more than one-third of a generation fails to enter the career path it expected.

04The Multiplier: The Pyramid Model Breaks First

Why it matters

The industries doing the hiring are also the industries whose economics depend on it. Law firms, consulting firms, and accounting firms have run on a pyramid: a large base of junior labor billed out at high margins subsidizes senior partner compensation. AI is removing the routine drafting, research, and analysis work that justified hiring large junior classes — which means the pyramid itself, not just the entry-level job count, is the thing breaking.

Source: PwC, Accenture, and McKinsey public disclosures and reporting via Innovaiden and Cornford & Cross industry analyses, 2026; Law360 Pulse March 2026 survey.

This adds a segment the original stock screen did not cover. Accenture (ACN) is the cleanest public proxy for the transition — and a genuinely mixed case rather than a clean loser, since it is shrinking its traditional headcount while its AI/data-services headcount is roughly doubling. Staffing and recruiting companies (Robert Half, ManpowerGroup) are directly exposed to entry-level and contingent hiring volume, independent of which sector it is in. Legal-tech and e-discovery platforms sit on the other side of the same trade: they benefit from the same automation that is shrinking associate classes.

05The Spending Shift

Setting up the chart

White-collar workers are roughly 63% of employed 20–34 year-olds but currently generate about 70% of this group's modeled spending, because of higher average earnings. As career-track hiring narrows, that imbalance unwinds.

This is an illustrative, static translation, not a demographic forecast — it does not yet incorporate population growth, inflation, wage growth by occupation, or new occupations AI itself may create. Its value is direction and order of magnitude, not a precise 2036 dollar figure.

06What This Consumer Already Buys Differently

Setting up the chart

The Consumer Expenditure Survey already shows how blue-collar and service households allocate a materially different share of their budget than professional households — this is observed data, not projection, which is what makes it the most load-bearing evidence in the whole thesis.

Source: Bureau of Labor Statistics, Consumer Expenditure Survey, 2024 averages by household occupation of reference person.

The clearest illustration is the vehicle economy. Blue-collar / service households spend less than half as much as professional households on new vehicles — but nearly as much on fuel, insurance, maintenance, and repair.

The transition is not from expensive cars to cheap cars. It is from buying the next asset to keeping the existing asset working — a pattern that can persist even if aggregate transportation demand holds steady, and one that favors repair and maintenance businesses over vehicle manufacturers.

07The Feedback Loop: Political Risk to the AI Trade Itself

Why it matters

Most versions of this thesis treat the AI infrastructure trade and the AI consumer-shift trade as unrelated — one bullish, one a downstream implication. They are not unrelated. A large, credentialed cohort that cannot find the work it expected is close to the textbook definition of what political scientists call "elite overproduction," a pattern historically associated with rising political instability independent of the headline unemployment rate. That instability, if it materializes, is most likely to land as regulation or taxation on the AI industry itself — the same companies the first-order AI trade is built on.

This is not hypothetical sentiment. A King's College London poll found that a third of UK university students believe AI-driven job losses could trigger civil unrest. The practical implication for portfolio construction: a book built entirely long the first-order AI trade and long the second-order consumer-shift winners is not fully hedged against the scenario this report describes — a modest short or reduced-conviction position in AI infrastructure names most exposed to regulatory backlash is a natural complement, not a contradiction, of the thesis.

08Other Multi-Year Effects to Watch

These effects compound over a longer horizon than the 2036 window modeled above. None is likely to be a near-term stock driver on its own, but each shapes which companies are durable beneficiaries versus which are riding a temporary mix shift.

09Where the Money Moves: Winners and Losers

The following are preliminary structural exposures, not valuation calls — a structural winner can still be a poor investment if it is overvalued, over-levered, or poorly executed. Section 10 provides the framework for making that distinction.

Strongest structural winners

Strongest structural losers

10The Screening Framework

Structural exposure alone is not an investment thesis. The framework below converts exposure into a repeatable score, then combines it with valuation to identify where the market has not yet caught up.

The four-box test

Mispricing score components

11Risks and What Would Make This Wrong

The severe scenario modeled in this report is a stress test, not a base-case prediction. The table below combines the original risk set with the competing explanations surfaced by checking the thesis against current data — several of which change how the model should be built, not just whether the thesis survives.

12Conclusion

The first-order AI investment has been infrastructure. The second-order investment is the consumer that infrastructure leaves behind — and, per Section 7, that second-order shift carries a third-order political risk back to the first-order trade.