Computer Science Graduate Unemployment Hit 7.0%. The Data Cannot Blame AI

Recent computer science graduates had a 7.0% unemployment rate in the New York Fed's latest table by major, against 4.2% across all majors. Computer engineering reached 7.8%, second of 73 majors. Computer science graduate unemployment is real and it rose between the last two releases. The table cannot say why, and it was never built to.

The short answer

Yes, recent computer science graduates face above-average unemployment: 7.0% against 4.2% for all majors in the New York Fed's table built on 2024 Census data. They also earn more and are less often underemployed. The table measures outcomes for 22 to 27 year olds. It does not measure causes, so it cannot attribute any of the gap to AI.

What the New York Fed's recent graduates by major table measures

The table is a one-year snapshot from a household survey. The New York Fed's Labor Market for Recent College Graduates page says the outcomes by major data "encompass only the most recent year period available" from the American Community Survey. It is updated annually in February.

Three limits follow from that design. The file carries one year of data per major and no history. It covers early-career graduates aged 22 to 27, which spans several graduating classes rather than one. And it records a person's degree field, not the job they were seeking when the survey reached them.

The downloadable file lists 73 majors plus an overall row. Each row has five numbers: unemployment, underemployment, early-career median wage, mid-career median wage and the share holding a graduate degree. The file publishes no sample sizes and no standard errors, and it does not state its data year. The 2024 reference year comes from NewsNation's coverage of the February 2026 release.

Two surveys, two recent-graduate rates

The 4.2% all-majors figure will look wrong if you follow this topic. The Fed's quarterly headline series put recent graduates at 5.2% in the third quarter of 2026, its lowest level in more than a year. The two numbers come from different series, covering different periods.

Coverage that compares computer science's 7.0% with a quarterly 5.6% understates the gap, because the like-for-like figure inside the same table is 4.2%. The quarterly series is the better guide to what is happening now, but it has no breakdown by major.

Computer science graduate unemployment across Fed releases

Computer science graduate unemployment rose across the two most recent releases I could verify, and so did the major's rank. It went from 6.1% to 7.0%, moving from seventh to fourth. Computer engineering went from 7.5% to 7.8%, moving from third to second.

2023 column as reported by CNBC, 16 May 2025, to one decimal or less. 2024 column from the New York Fed CSV, retrieved 5 November 2026. Changes are Zan Digital calculations. The 2023 all-majors row in the table was not verified, so it is omitted rather than estimated.

Look at the third row before you draw a conclusion from the second. Art history more than doubled in one release, a bigger move than either computing major. I know of no argument that AI suddenly hit art history graduates in a single year. Large annual swings are what thin samples by major can produce.

Why earlier releases are not charted

The Fed's page says the by-major file shows only the latest year. It replaces last year's table instead of extending it, so no official multi-year series by major exists to download.

I could not trace any 2022 figure for computer science or computer engineering to a Fed document I could open, so none appears above. The one earlier reading I can stand behind is broader. EIG's analysis of the ACS found the six majors in its computer and information sciences category rising from just under 4% in 2022 to just under 6% in 2023.

The direction holds even if the rank does not

A rise in the broad group, followed by two elevated annual tables for the narrow majors, is more convincing than any single row. My read is that the elevation is genuine and the precise ranking is not.

Why the computer engineering rank is mostly noise

The rank is the least reliable number in this table, yet it is the one in the headlines. Connor O'Brien at the Economic Innovation Group rebuilt the 2023 estimates with survey replicate weights. His 95% interval for computer engineering's 7.5% ran from 4% to 11%. Physics ran from 2% to 12%, and public policy from 0% to 13%.

Eleven places separate second from thirteenth, yet only 1.68 points of unemployment do. That spread sits comfortably inside a 7-point interval, assuming the 2024 margin resembles the 2023 one.

Coverage says computer science ranks fifth. The CSV's three decimals put it fourth, at 6.992 against 6.95 for performing arts. That difference of 0.042 points decides the placing, and I would not quote the placing at all.

I would not say computer engineering graduates are worse off than anthropology graduates either. The 7.783 versus 7.922 gap means nothing statistically. The fair summary is that a group of majors, computing among them, sits clearly above the 4.2% average in the latest table.

Entry-level software pay and underemployment complicate the picture

Computing graduates who do land a job fare far better than the unemployment rank implies. The same CSV that ranks computer science fourth on unemployment shows a median early-career wage of $87,000, against $58,000 overall. Computer engineering shows $90,000. Those are 50% and 55% above the all-majors median.

Underemployment points the same way. The Fed counts someone as underemployed when they work in a job that does not typically need a degree. Computer science is at 19.1% and computer engineering at 15.8%, against 39.4% overall. Fine arts is at 58.9%, with unemployment close to computer science.

An unemployment-adjusted wage

To put the two columns together I multiplied each major's median wage by its employment share. It is a crude measure, since a median is not an average and the wage only describes people who have work. It does answer one simple question: which major looks better once you price in the risk of no job?

Computer science comes out about 46% above the all-majors figure, and ahead of electrical engineering, which has less than half its unemployment. Median wages also rose against the prior table. CNBC reported $80,000 for these majors in the 2023 data, so unemployment and pay rose together. That combination fits a market that filters harder, though it does not prove one. Rising medians can also mean lower-paid candidates dropped out of the employed group.

The BLS test: computing payrolls fell, but less than the table implies

BLS data confirms weaker demand for computing labour, and it shows nothing close to a collapse. No BLS series reports unemployment by college major, so these checks are indirect.

Start with the payroll count. BLS employment in computer systems design and related services peaked at 2,483,500 in March 2023 and stood at 2,355,400 in September 2026. That is a fall of 128,100 jobs, or 5.2%. The same line is still 9.4% above January 2019, and it covers only one industry, since many software workers are employed elsewhere.

Next, unemployment among people with computing jobs. In CPS Table 25, the rate for computer and mathematical occupations rose from 2.8% in 2024 to 3.3% in 2025. BLS assigns unemployed people to an occupation by their last job, so a graduate who never held a computing role does not land in this row. A low rate here alongside a high rate in the Fed table fits a problem concentrated at the point of entry, not across the profession.

Finally, the closest thing BLS has to a graduate measure. Its October supplement found 15.3% of recent bachelor's recipients aged 20 to 29 unemployed in October 2024, a rate it said had changed little from the previous year. That group earned degrees that same year, which is why the number is so much higher. It has no field-of-study split. BLS also notes that no CPS data were collected for October 2025 because of the shutdown.

BLS's Occupational Outlook Handbook still projects software developer employment to grow 10% from 2025 to 2035, while computer programmers are projected to decline 7%. Those are projections, not results. The occupational split behind them is covered in the breakdown of software developer employment in BLS data.

The NACE test: employers still plan to hire CS graduates

NACE's employer surveys point the opposite way from the unemployment rank. In its Winter 2026 Salary Survey, at least 60% of responding employers planned to hire computer science majors from the class of 2026. Those employers projected a bachelor's average of $81,535, up 6.9%, the largest increase NACE reported for any major group.

Read the sample before you lean on that. It rests on 150 returned surveys, a 20.3% response rate, all from NACE employer members. Those are firms running organised campus recruiting, which tells you what they intend and nothing about employers who stopped showing up. The wider caveats on NACE's plans are set out in the reconciliation of conflicting entry-level hiring data.

The realised-outcome data is no sharper for this question. NACE's first destination report found 85.7% of class of 2024 bachelor's graduates employed or in further study within six months, from 363 institutions and about 571,500 graduates. Its public article prints no computer science figure.

Both sources can be right at once. Employers can keep hiring computing majors, at rising pay, while a much larger graduate pool produces more people who do not land those roles.

What the NY Fed data cannot attribute to AI

The by-major table cannot attribute any of this rise to AI, because it collects nothing about causes. Respondents report their degree and their employment status. Nothing in it records why someone was not hired, whether the job existed, or whether software reduced demand for it.

At least four explanations fit the same rows, and they overlap in time. ChatGPT launched on 30 November 2022, and computer systems design payrolls peaked four months later.

Supply is the row most coverage skips. Computing fields went from 2.8% of all bachelor's degrees in 2012-13 to 5.4% in 2021-22, so the classes now inside the Fed's age window are far larger than those before them. The overhiring row has its own test against company hiring histories, in the analysis of whether AI layoffs are overhiring corrections.

Why the pattern across engineering does not settle it

The tempting test is to compare computing with other engineering. Electrical engineering sits at 3.2% and mechanical engineering at 4.4%, so technical training alone is not the problem. Computer science runs 3.8 points above electrical engineering, which is a real signal about software labour specifically.

The trouble is that every story predicts the same gap. Emanuel, Harrington and Pallais use software engineering and mechanical engineering as their examples of remotable and non-remotable work. Software work is also widely treated as exposed to AI, and the payroll line above shows its own boom and retreat, up 15.3% from January 2019 to the March 2023 peak. The comparison confirms that software is different. It does not say which difference matters.

Those authors argue that the rise in youth unemployment predates the rapid diffusion of AI, and they find age gaps persist even with AI exposure held constant. They also concede that AI "may play a more primary role" going forward. The competing method is assessed in the comparison of the Stanford and Yale AI jobs studies. I would not cite either study as settling computer science specifically, because neither the remote-work post nor the Stanford update reports results for that degree.

Where this reading of the CS grads job market is weakest

My position is that the elevation is real and the cause is unproven.

Two releases are not a trend

I set out to chart this across many releases and verified two, plus one broader series. A reader could fairly say I have too little history to call anything a rise. The Fed's replace-each-February format is the reason, and an archive of old files could overturn what I describe here.

The data is also old. A table on 2024 data says nothing about the class of 2025 or 2026. The quarterly series has since improved, from 5.9% in October 2025 to 5.2% in September 2026, but it covers all majors, so computing could be moving either way inside it.

The case that AI is already in these numbers

The strongest objection comes from the Stanford Digital Economy Lab. It reports the gap for young workers in exposed occupations widening from 15% to 19% in a year. It also says the divergence survives excluding technology firms and computer occupations, and controlling for interest rates and remote work. That is hard to dismiss as a tech-cycle hangover.

If that reading is right, "the data cannot attribute it" could become a comfortable way to ignore a real shift, and new evidence would arrive years late. Postings data, covered in the piece on the junior developer pipeline, points the same way. I still think "AI did this" outruns the by-major table, but I hold that view less firmly for 2026 than for 2023.

Frequently asked questions

Where to start if you set graduate intake

Stop using this table as an input to your hiring plan, in either direction. It does not show that graduate engineers have become cheap or plentiful at the quality you need, and it does not show that AI has made them unnecessary. Pull your own numbers instead: applications per opening, offer acceptance and first-year retention for each intake since 2023. If those have not moved, the 7.0% is somebody else's problem.

Then set a reminder for February 2027 and save the Fed's CSV the day it updates, because the current file will be replaced. Check one number first: the gap between computer science and electrical engineering, now 3.8 points. If it widens while computer systems design payrolls stabilise, the overhiring story weakens and something structural deserves more weight.

Related reading