College grads shut out of AI-exposed majors since 2022 are ending up in retail and food service instead of the white-collar jobs they studied for

Census research found employment among workers aged 22-24 in the most AI-exposed industries fell 12% in the 10 quarters after ChatGPT's late-2022 release, while graduates from the most AI-exposed majors saw initial employment odds decline 5 percentage points and earnings fall 13%. Roughly half of the earnings loss reflected lower pay within hiring industries, with the remainder driven by shifts into lower-wage retail and restaurant jobs. Computer-science enrollment fell 8.4% year over year in spring 2026, while recent-college-graduate underemployment reached 42% in Q2 2026.

Analysis

The investable read-through is not broad consumer weakness; the affected cohort is too small to move aggregate demand materially. It is a labor-share transfer within knowledge work: platforms and large enterprises can convert fewer junior hires into structurally lower SG&A and higher revenue per employee, favoring MSFT, GOOGL, META and ORCL over labor-arbitrage models such as ACN, EPAM and GLOB. The second-order risk is reduced apprenticeship capacity: if firms underhire juniors for multiple years, scarce senior engineers could regain pricing power in 2028-30 and constrain product execution, partially offsetting near-term margin gains.

The delayed enrollment response creates a potentially important 6-18 month supply effect. A smaller pipeline of computer-science graduates will not matter while AI still substitutes for routine development, but it raises the odds that AI infrastructure, cybersecurity and semiconductor design face specialized-talent bottlenecks once enterprise deployment shifts from experimentation to production. That is more supportive of high-value engineering vendors and design-tool ecosystems (CDNS, SNPS) than of generic IT-services providers whose historical model depends on pyramiding junior labor.

Near term, this is primarily a margin and multiple-dispersion theme rather than a new earnings catalyst: investors already expect AI productivity gains at mega-cap platforms. The more underappreciated downside is for staffing and consulting estimates if client demand increasingly requires smaller project teams; quarterly utilization, headcount growth and bookings at ACN/EPAM/GLOB are the cleanest 1-3 month falsification signals. GS is not a direct operational beneficiary; the enrollment data are useful for its research franchise but immaterial to earnings.

Key Decisions for Investors

• Establish a 6-12 month pair: long MSFT / short ACN, sized beta-neutral. MSFT has multiple AI monetization paths while ACN faces both reduced billable headcount per project and slower campus-hiring leverage; reassess if ACN organic bookings accelerate and utilization improves for two consecutive quarters.
• Add a 6-18 month long basket in CDNS and SNPS on sector pullbacks rather than chasing headline strength. Specialized design workflows are less exposed to junior-code substitution and could benefit if constrained engineering supply raises customer demand for productivity tools; exit if semiconductor-design R&D budgets materially decelerate.
• Maintain an underweight/watch on EPAM and GLOB rather than a new outright short until next earnings. Upgrade-to-short only if management guides to falling headcount or weaker digital-engineering demand while AI-related pricing remains flat; the key risk is AI creating incremental modernization projects that outweighs labor substitution.
• Do not trade GS on this development. Monitor its quarterly compensation ratio and technology-investment commentary only as a secondary indicator of whether financial-services firms are converting AI labor savings into margins rather than redeploying them into revenue-generating roles.