AI job fears vs. data: what the newest research actually says

On 25 September, CNBC published the latest reporting in this dossier on how AI is reshaping early-career hiring. The headline numbers are uncomfortable for graduates. They do not support the simplest version of the story.

According to the Federal Reserve Bank of New York, cited by CNBC, college graduates aged 22 to 27 had an unemployment rate of 5.7% as of June 2026. That is above the national average and a full percentage point higher than the same cohort two years earlier. For all young workers, the rate was 7.2% in June. The underemployment rate, meaning graduates working in jobs that do not require a bachelor's degree, reached 42% for recent graduates, against 33.7% for all college graduates. Those figures describe a weak entry-level market. On their own, they do not prove AI caused it.

The New York Fed's own study of job postings is explicit on that point.

"While AI may be contributing to recent labor market developments, it is not the main driver of the slowdown in hiring," the NY Fed wrote, according to CNBC. Its employer surveys found that firms "mostly intend to incorporate AI mainly via retraining, with limited effects on hiring." That is a much narrower claim than the one circulating in commentary. It also comes from the same institution whose graduate unemployment numbers are being used to argue the opposite case.

What is clearly changing is the hiring process itself. LinkedIn research cited by CNBC found that 66% of recruiters said at the start of 2026 they planned to increase their use of AI for pre-screening interviews, while 81% of job seekers said they had used or planned to use AI in their search. Both sides of the transaction are now partly automated. That makes it harder to read any single hiring statistic as a clean measure of demand.

Skills data points the other way

The evidence on skills is moving in a different direction from the displacement narrative. Handshake, an early-career platform that has worked with OpenAI on student tech tools, says graduating seniors in 2026 mentioned AI skills on their resumes at twice the rate of the class of 2022. Its analysis found 74% of those mentions were tied to real-world projects rather than coursework. "The data tells us that early talent isn't being displaced by AI, they're being amplified by it," Handshake wrote in a blog post cited by CNBC.

The National Association of Colleges and Employers reports a similar shift in demand. AI skills appeared in 16.5% of job descriptions in the spring, up from 10.5% the previous autumn. Overall, 28% of employers say they are seeking early-career talent who can use AI at work. NACE describes a "striking disconnect" between some Gen Z candidates and the labour market. Employers ask for AI readiness, while a significant share of students question whether AI belongs in their job search at all.

The blue-collar side of the same boom

AI is also creating demand in trades that cannot be automated or offshored. CNBC reported on 26 September that welders, plumbers, HVAC technicians and electricians are in high demand for data centre construction, though many of those jobs are temporary. Maria Flynn, president and CEO of Jobs for the Future, told CNBC that data centres are only one piece of a larger puzzle that includes grid modernisation and transport upgrades.

The pay data is striking. Nicole Bachaud, a labour economist at ZipRecruiter, told CNBC that the mean minimum salary for data centre jobs rose 125.1% year over year to nearly $208,000. She attributed the rise to highly specialised engineering roles pulling the average up. Postings for welders and pipefitters are up 164% year over year, according to Bachaud's data. Apprentice-level technicians can take home $40,000 to $60,000, Flynn said, with experienced electricians above $100,000.

Justin Sinkovich of Columbia College Chicago pointed CNBC to Louisiana, where Amazon committed $12 billion to a data centre with 540 on-site jobs, including 1,700 electricians, technicians and security personnel, and Meta's Hyperion project is valued at $27 billion. Those roles, he noted, cannot be done remotely. The risk to this part of the story is political. CNBC reported that states including New York and Texas have moved to slow development, and local freezes have followed public backlash. That could hit the labour market trend it just created.

An academic paper posted on 22 September to the NBER conference site, titled Artificial Intelligence and Labor Market Reallocation, sits alongside these findings without settling them. The dossier contains the PDF but not a readable summary of its conclusions, so its contribution cannot be fairly characterised here.

Read together, the newest sources point to reallocation rather than collapse: weaker entry-level hiring, stronger demand for AI skills and for trades, and employers planning to retrain rather than replace. The disagreement between the NY Fed's "not the main driver" finding and the graduate unemployment numbers is real. Another round of commentary will not resolve it.