Is a STEM Degree Still Worth It? The Best-Paid Major Also Has the Second-Worst Unemployment
The pitch: study STEM and you’ll be fine.
The test: computer engineering graduates face 7.8% unemployment — the second-highest rate among the 74 majors the New York Fed tracks. Those same graduates earn a median of $90,000 in their first years out of school, the highest early-career wage on that list.
Same table. Same release. Same year.
So which is it? That question is worth about $100,000 of tuition and four years you don’t get back, so let’s run it properly — because “STEM” is being sold to you as one product, and the data says you’re choosing among at least three.
What are you actually being sold?
The New York Fed’s labor market data for recent college graduates, updated February 4, 2026 from Census American Community Survey microdata, reports two different things about every major, and the gap between them is where the money hides.
Unemployment is the share of graduates who can’t find a job at all. Underemployment is the share working a job that never required the degree you financed.
| Major | Unemployment | Underemployment | Early-career wage | Share with grad degree |
|---|---|---|---|---|
| Computer Engineering | 7.8% | 15.8% | $90,000 | 39% |
| Computer Science | 7.0% | 19.1% | $87,000 | 33% |
| Physics | 6.6% | 29.1% | $67,000 | 67% |
| Information Systems | 6.0% | 25.6% | $67,000 | 28% |
| Mathematics | 5.8% | 26.2% | $70,000 | 51% |
| Chemical Engineering | 4.7% | 17.9% | $85,000 | 48% |
| Mechanical Engineering | 4.4% | 20.1% | $80,000 | 39% |
| Biology | 4.3% | 51.1% | $45,000 | 64% |
| Chemistry | 4.3% | 42.8% | $50,000 | 67% |
| Electrical Engineering | 3.2% | 21.1% | $82,000 | 48% |
| Civil Engineering | 2.3% | 15.6% | $75,000 | 37% |
| Aerospace Engineering | 2.2% | 14.7% | $85,000 | 46% |
| Nursing | 2.1% | 12.8% | $70,000 | 30% |
For scale: the same source put unemployment for all recent graduates at 5.7% in June 2026, against 4.1% for all workers and 2.9% for college graduates of every age.
Read the unemployment column and computing is the worst buy on the page. Read the underemployment column and it’s one of the best. Anyone quoting you one column without the other is selling, not advising.
Which failure are you actually buying?
These two columns describe opposite ways to lose money, and they demand opposite hedges.
A computer science graduate’s risk is not getting in the door. Clear it and the purchase performs: 19.1% underemployment is far below the 42.9% median across all 74 majors, and $87,000 to start is second only to computer engineering.
A biology graduate’s risk runs the other way. Only 4.3% are unemployed — but 51.1% are underemployed, at a $45,000 median. Biology graduates get hired. Roughly half don’t get hired as scientists.
One is a toll booth at the entrance with clear road behind it. The other is an open entrance onto a road that doesn’t go where the brochure implied. You cannot hedge both with the same plan.
Is computing still worth the tuition?
Steelman first, because it’s strong. Computing has the highest early-career pay in the dataset, underemployment near the bottom, and — per our own analysis behind the computer science worth-it breakdown — a median of $114,482 ten years after entry against roughly $19,559 in median debt at graduation. That is a debt load of about a sixth of one year’s pay at the ten-year mark. On payback period alone, it is one of the best deals in American higher education, and nothing below changes that.
Now the test. What exactly got worse, and is it permanent?
The sharpest evidence comes from splitting one industry in half. IEEE Spectrum, citing Bureau of Labor Statistics figures, reported in December 2025 that between 2023 and 2025, US employment for programmers fell 27.5% — while employment for software developers fell 0.3%.
Same industry, same tools, a 27-point gap. The difference is the job description. Writing code to someone else’s specification is what current AI does well. Deciding what to build, for whom, and whether the output is actually correct is what it doesn’t.
The general version comes from Stanford’s Digital Economy Lab. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen’s paper “Canaries in the Coal Mine?”, built on records from the largest US payroll processor, found a 16% relative decline in employment for workers aged 22 to 25 in AI-exposed occupations, measured against both older workers in the same roles and workers in less-exposed fields. The declines concentrate, in their words, “in occupations where AI is more likely to automate, rather than augment, human labor.”
The verdict on this tier: you are not buying a worse degree. You are buying the same degree with a harder first year. That is a real cost — a delayed start compounds across a career — but it is a cost at the entrance, not a markdown on the asset.
What are you actually paying for with a license?
Look at the bottom of the unemployment column: nursing 2.1%, aerospace engineering 2.2%, civil engineering 2.3%. Now check their underemployment — 12.8%, 14.7%, 15.6%, the three lowest in the entire 74-major dataset.
Between them, these fields draw on two protections that current AI handles badly. Nursing and civil engineering have both; aerospace leans on the second.
A legal credential. A registered nurse and a professional engineer hold a license a model cannot hold. That barrier is annoying to clear and precisely why it holds wages up once you have.
A physical worksite. Bridges, patients, airframes and process plants are not text. AI compresses the paperwork around that work. It does not do the work.
Price nursing honestly and it is the clearest cost-benefit case on the page: a $70,000 early-career median, the lowest underemployment of any major in the dataset at 12.8%, and — from our data — a $17,310 median debt at graduation, among the lowest of any STEM field we track. That is roughly a quarter of one year’s starting pay in debt. What you give up is upside: registered nurses are projected to grow +4.9% through 2034, less than a third of computing’s rate, and nursing’s ten-year median of $89,753 sits well below computer science’s. Low ceiling, high floor. Right now the floor is the scarce commodity. Our nursing worth-it analysis and nursing path comparison price the ADN-versus-BSN version.
The engineering disciplines are the same trade in a higher bracket: civil and mechanical engineering reach $106,672 and $119,489 at ten years, with entry unemployment under half of computing’s.
Is the bachelor’s the product, or just the down payment?
This is the tier where the pitch and the product diverge most, and almost nobody prices it correctly.
Biology: 51.1% underemployment, $45,000 early career. Chemistry: 42.8% and $50,000. Both under 5% unemployment — they find work, just not in a lab.
The graduate-degree column tells you why. 64% of biology majors and 67% of chemistry majors eventually hold a graduate degree, against 30% for nursing and 33% for computer science. In these fields the bachelor’s is not the credential that pays. It is the prerequisite for the one that does.
That produces a two-stage earnings curve. Biology starts at a $45,000 median and reaches $82,087 at ten years in our data — real growth, mostly routed through medical school, a PhD or a professional program. Physics runs the same shape: $67,000 early, $105,505 at ten years, with 67% holding a graduate degree.
This is not an argument against these majors. It is an argument against a four-year budget for an eight-year plan. If the credential that pays arrives in year eight, then years five through eight are part of the purchase price, and the opportunity cost of those years belongs in the calculation too. Our ROI calculator will take the full timeline.
Does the entry-level number actually price the degree?
Put the two datasets side by side, because a single year is a terrible way to price a forty-year asset.
| Major | Early-career (NY Fed) | 10 years in (Census PSEO) |
|---|---|---|
| Computer Science | $87,000 | $114,482 |
| Nursing | $70,000 | $89,753 |
| Civil Engineering | $75,000 | $106,672 |
| Biology | $45,000 | $82,087 |
| Physics | $67,000 | $105,505 |
Every one still climbs. What AI has done so far is narrow the first step in the fields most exposed to it — not shorten the staircase.
That distinction should change your behavior without changing your decision. A hiring market four years from now is not the one in today’s headlines, and if you pick your major off a two-year panic, you have optimized a forty-year asset against a number that will have moved before you graduate.
What’s the strongest case against everything above?
Two pieces of evidence cut against this article’s own thesis, and leaving them out would be exactly the kind of selective quoting it warns you about.
Most employers say they aren’t doing what you think. National Association of Colleges and Employers data, reported by IEEE Spectrum, found 61% of employers say they are not replacing entry-level jobs with AI, while 41% are discussing or planning to augment those roles over the next five years. The dominant posture is augmentation.
The Stanford authors revised their own timing. In a February 2026 follow-up, Brynjolfsson and his co-authors applied firm-time fixed effects and found the employment decline is statistically significant only after 2024 — not from late 2022, as the original framing suggested. They ruled out interest rates as the primary driver, noting AI-exposed jobs are on average less rate-sensitive. But they wrote plainly: “we do not believe that AI is always and everywhere the sole determinant of employment,” and called it “an active area for scientific research.”
The honest reading: a tech-sector correction, a rate cycle and an AI capability jump landed on the same graduating cohort inside eighteen months, and the shares are genuinely unsettled. Anyone selling you a clean causal story about your major — in either direction — is selling something.
How do you test this before you sign?
- Make them answer both questions. “What share of your graduates are employed?” and “employed in the field?” are different questions. For biology and chemistry the gap between them is the entire decision. A department that can only answer the first has answered the second.
- If you want computing, budget for the first job, not just the degree. The ceiling is intact; the door narrowed. Internships, shipped work and co-op placements are worth more in this specific market than a GPA decimal. Our computer science path comparison runs associate versus bachelor’s on cost and time.
- Move up the automate/augment line inside your own field. The 27-point gap between programmers and software developers is the same gap between lab technician and lab scientist, between drafter and design engineer. Specification-following work is exposed. Judgment, verification and accountability are not — yet.
- Price the graduate degree when the field requires one. At 64–67% grad-degree share, biology, chemistry and physics are multi-stage purchases. Two extra years of tuition plus forgone earnings moves an ROI verdict far more than a $5,000 difference in starting salary.
- Treat licensure and physical work as financial instruments. The three lowest underemployment rates in the dataset — nursing, aerospace and civil engineering — sit behind either a legal credential or a worksite a model cannot enter. That is not a coincidence, and it is the most durable protection currently visible in the data.
- Buy the uncertain years cheaply. Two years of transferable math and science costs the same at community college prices in every one of these tiers, and buys time to watch a market that is honestly unsettled. Our transfer pathway data covers what actually moves.
- Don’t confuse the field with the credential. If it’s an AI-specific degree you’re weighing, we tested that separately in is an AI major worth it — the short version is that 141 students in the entire country earned one last year.
The verdict
Worth it if you pick the tier deliberately and budget for how that specific tier fails. Every field in the table above clears the $82,500 mid-career median across all 74 majors in the New York Fed data — biology only barely — and the two with the worst entry unemployment also carry the highest pay and among the lowest underemployment. On a payback-period basis these remain some of the strongest purchases available to an undergraduate.
Reconsider the framing if you were sold “STEM” as the answer itself. Computer engineering and biology are both STEM. One has 15.8% underemployment at $90,000; the other has 51.1% at $45,000. That is not a difference of degree, and no single piece of advice covers both.
What AI changed, on the evidence available in August 2026, is the price of the first rung in fields where the entry-level task is automatable text. What it did not change: technical degrees still pay, licensed and physical work is holding, and the fields that always required a graduate degree still require one.
Pick the tier. Then price how that tier fails, not how the brochure says it succeeds.
Sources
- Federal Reserve Bank of New York — “The Labor Market for Recent College Graduates,” outcomes-by-major table (U.S. Census Bureau, American Community Survey via IPUMS; U.S. Department of Labor, O*NET) — updated February 4, 2026 — https://www.newyorkfed.org/research/college-labor-market
- Federal Reserve Bank of New York — “The Labor Market for Recent College Graduates,” unemployment series (Current Population Survey via IPUMS) — updated August 6, 2026 — https://www.newyorkfed.org/research/college-labor-market
- Erik Brynjolfsson, Bharat Chandar and Ruyu Chen — Stanford Digital Economy Lab — “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence” — November 13, 2025, updated February 9, 2026 — https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- Erik Brynjolfsson, Bharat Chandar and Ruyu Chen — Stanford Digital Economy Lab — “Canaries, Interest Rates, and Timing: More on Recent Drivers of Employment Changes for Young Workers” — February 9, 2026 — https://digitaleconomy.stanford.edu/news/canaries-interest-rates-and-timinga-more-on-recent-drivers-of-employment-changes-for-young-workers
- Gwendolyn Rak — IEEE Spectrum — “How AI Is Reshaping Entry-Level Tech Jobs” (BLS employment figures; NACE Job Outlook 2026 survey) — December 25, 2025 — https://spectrum.ieee.org/ai-effect-entry-level-jobs
- Michael T. Nietzel — Forbes — “Unemployment And Underemployment Rates Among Recent College Graduates” — February 23, 2026 — https://www.forbes.com/sites/michaeltnietzel/2026/02/23/unemployment-and-underemployment-rates-among-recent-college-graduates/
- U.S. Census Bureau — Post-Secondary Employment Outcomes (10-year median earnings by field of study), and U.S. Department of Education College Scorecard (median debt at graduation), analyzed in-house — https://census.gov/data/experimental-data-products/post-secondary-employment-outcomes.html
- U.S. Bureau of Labor Statistics — Employment Projections, 2024–34 (occupation-level growth rates mapped to fields of study) — https://www.bls.gov/emp/
