New graduates from the most AI-exposed degrees are losing ground at rates that mirror a recession, and the newest response from universities is a hard pivot toward factories that make the chips AI runs on. A US Census Bureau working paper published in September 2026 — Graduating into Disruption — puts numbers on the first years of the generative AI era for people who graduated into it. The same week, CNBC reporting picked up by the Times of India described American universities redirecting students toward semiconductor manufacturing careers.
🔍 THE BOTTOM LINE: For now, AI is squeezing graduates whose jobs most resemble what AI does — writing, coding, analysing — and the labour market’s safety valve is physical work the technology cannot yet do. The earnings data is sobering; the pivot story says the adjustment is already under way.
What the Census Economists Actually Measured
The paper, by Cody Orr, Lee C. Tucker and Lawrence Warren at the Census Bureau’s Center for Economic Studies, links administrative employment records to college majors, then ranks majors by how AI-exposed their typical first occupations are. The Register’s summary of the paper, published on 18 September 2026, is blunt: “Slinging coffee and waiting tables ain’t just for liberal arts grads any more.”
The headline findings for the most exposed decile of majors — which includes computer science, mathematics, statistics, some engineering disciplines and, the authors note, journalism:
- Likelihood of initial employment: down five percentage points.
- Full-quarter initial earnings: down 13 percent.
- Roughly half of the earnings decline reflects a shift into lower-wage sectors such as restaurants and retail.
The authors compare the earnings hit to “the earnings losses associated with graduating into a large recession.” Graduates in less exposed fields — nursing and many education majors — have not seen the same deterioration.
The Nuance Most Coverage Skips
This is a working paper, not a final verdict. It has not undergone the review accorded Census Bureau publications, and the authors are explicit that their estimates identify a pattern, not a single mechanism. Firms may be hiring fewer graduates because AI exists — even where the firms do not use AI themselves, because they expect disruption or doubt AI can reliably filter candidates.
The economists also tested the obvious alternative explanations. Oversupply of graduates? They found the opposite pattern: growth in graduates in AI-exposed majors levelled off quickly after ChatGPT’s release, and some department enrolments have started to decline. Self-employment or grad school absorbing the shift? Their data suggests those patterns look more like weak labour demand than opportunity. They told the Register they “hesitate to extrapolate this short-term trend into a long-term doomsday forecast” — and the paper notes there is little evidence so far of long-term scarring, with effects easing as graduates move further from graduation.
Where the Jobs Went Instead
The second half of the story is the labour market adjusting in real time. As CNBC’s report describes, universities including Purdue and Arizona State are investing heavily in preparing students for semiconductor manufacturing — a sector where McKinsey and the SEMI Foundation project a US shortage of up to 157,000 workers by 2030.
The economics are concrete: Samsung is bringing its first Texas fab online as part of a $35 billion investment expected to create roughly 3,500 jobs; US chip jobs typically pay between $127,000 and $187,000, with senior roles exceeding $238,000, according to the SEMI Foundation. Jon Taylor, executive vice president of Samsung’s semiconductor division in Austin, told CNBC: “We’re hiring engineers, we’re hiring technicians, we’re hiring people in supply chain… it’s a bit of a race against time right now.”
For students who watched software postings thin out, the semiconductor route is an unexpected alternative career path — same maths, different destination.
What This Means for New Zealand Graduates
New Zealand produces its own pipeline of computer science, mathematics and engineering graduates, and the direction of the US evidence matters for students choosing majors now. Two pieces of local context are worth holding alongside the Census data: SEEK’s August report showed NZ job ads rising for a 21st straight month with 4.0 percent of listings now asking for AI skills — demand that rewards exactly the synthesis-and-critical-thinking skillset this research suggests is becoming more valuable. And the Reserve Bank’s analysis of AI and the graduate first rung identified the same vulnerability the Census economists found: entry-level roles absorb AI’s earliest effects first.
The semiconductor angle has a NZ echo too. The data centre and trades boom shows the AI buildout creating physical-infrastructure jobs locally, a smaller-scale version of the American pattern where the jobs AI creates are the ones AI cannot do.
❓ FAQ
Which majors were most affected? The most AI-exposed decile included computer science, computer programming, mathematics, statistics, information sciences and some engineering disciplines — plus journalism. Nursing and education majors, whose work involves direct human interaction, did not show the same declines.
Does this prove AI is taking graduates’ jobs? Not by itself. The authors state the paper does not identify a single mechanism, and that firms might reduce graduate hiring for AI-adjacent reasons without using AI. But several alternative explanations — graduate oversupply, self-employment, grad school — did not fully account for the pattern.
What is the semiconductor opportunity? McKinsey and the SEMI Foundation project a US chip workforce shortage of up to 157,000 by 2030. Universities like Purdue and Arizona State are already building programmes, and US chip sector pay ranges from $127,000 to $187,000 with senior roles above $238,000, per the SEMI Foundation.
Is the earnings damage permanent? So far, effects attenuate as graduates move further from graduation, and the authors found little evidence of long-term scarring. The Great Recession comparison is about the size of the initial hit, not a confirmed permanent loss.
🔍 THE BOTTOM LINE
The Census paper gives the most rigorous picture yet of who pays the early costs of AI in the labour market: young people whose first jobs most resemble what AI does. But the same week’s hiring news shows the market’s response — graduates and universities moving toward work with an obvious physical floor. Career-wise, the lesson is less “avoid technology” and more that the pairing of technical skills with physical or human-centred work is where demand is landing first.
📰 Sources
- US Census Bureau CES Working Paper 26-56, “Graduating into Disruption” (September 2026)
- The Register, “Compsci grads facing recession-like job prospects thanks to AI” (18 September 2026)
- CNBC report via Times of India on universities pivoting to semiconductor careers (19 September 2026)
- SEMI Foundation / McKinsey semiconductor workforce projections
- SEEK NZ Employment Report August 2026
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
