Texas courted data centers for two years and is now turning them away. Applications from data centers and other large consumers to connect to the state grid rose from roughly 48 GW in 2023 to more than 474 GW today, according to a recent Reuters analysis, making Texas one of the fastest-growing homes for AI and cloud facilities. Last month Governor Greg Abbott, until recently one of the industry's strongest backers, announced a suspension of new data center interconnections. A new study supplies the grievance underneath the reversal: in the region that was supposed to gain most from the buildout, job postings in the fields most exposed to AI are down 8% and graduate employment is falling.
The researchers place the rise in graduate unemployment in the same window as the rapid spread of generative AI, and write in their published analysis that the pattern shows where the technology's effects on jobs and earnings will surface first. Their framing is deliberate. This is the place receiving the investment, not one of the places left out of it.
Abbott did not move because of the study. Public anger at data centers has been building, the midterms are close, and the governor began criticizing the AI industry before announcing the pause. Residents' objections run past the environmental complaints that follow these facilities everywhere to a plainer economic one: the technology inside the buildings may destroy jobs, and the buildings themselves give the local economy little.
The Texas finding is not alone. A similar analysis from Stanford's Digital Economy Lab found a marked decline in employment among young knowledge workers, alongside growth in occupations not considered vulnerable to AI automation, including manufacturing and retail. One of that report's authors said employment among people aged 22 to 25 is falling in the occupations most affected by AI and rising in the least exposed ones. Anthropic CEO Dario Amodei warned of this pattern earlier, with entry-level roles going first.
How many jobs AI has actually taken remains unclear. Some of the cuts are employers using the technology as a pretext for layoffs they wanted for other reasons. The uncertainty is doing damage on its own: in a Gallup survey run over the last three months of 2025, 72% of respondents said this is a "bad time" to look for a good job.
Two of the numbers in this story are doing very different work than they appear to. The 474 GW is a queue, not a load. Interconnection applications are cheap to file, the same project can sit in several of them, and nothing in that figure says how many megawatts will ever be energized. It is being read politically as though it were construction. The 8% is the opposite problem: a real measurement, but of postings rather than of people, and postings are the earliest and noisiest signal a labor market produces. The researchers are careful about causation, and so are the people who say some of the layoffs are cover for ordinary cost-cutting. A campaign season is not careful about causation.
What nobody in the reversal has explained is how pausing interconnections helps the graduates. The hiring the study measures is falling in occupations exposed to a technology that runs wherever the compute happens to sit. Substations in Texas are not what employs those workers, and sending the racks somewhere else does not put a single entry-level analyst back at a desk. The pause is a jobs answer to an energy question, or an energy answer to a jobs question; it is not clear which, and it works as politics either way.
That is the trap the state has walked into. Texas made itself the easiest place in the country to plug in AI infrastructure and received, along with the capital, the first clear local evidence that the technology thins out the entry-level end of the labor market. The facilities can be stopped at the state line. What gets trained inside them cannot.