All those construction jobs — do they disappear the day the data center opens?

Nine thousand people show up. The contract asks for 57.

Abilene, Texas is the cleanest example I could find, because it's the biggest. Oracle representatives told reporters the Stargate campus there created around 9,000 construction jobs and would leave about 1,000 long-term ones behind. Other local coverage put the long-term figure closer to 1,700. Those are the numbers that get repeated at council meetings.

Then there's the development agreement — the version with signatures on it. Reporting on that document says the minimum commitment is 57 full-time positions, averaging somewhere around $57,600 a year. The developer has said it expects more than a hundred. So depending on which piece of paper you're reading, the same project promises 57 jobs or 1,700.

I sat with that for a while. Both numbers can be technically true — one is a floor somebody agreed to be held to, the other is a forecast nobody signed. But only one of them is enforceable, and it isn't the big one. Abilene gave up 85% of the property taxes on that site for two decades to get it.

That's the part I'd want explained to me in plain words if it were my county, and I couldn't find a single place where anyone had.

The crew was mostly never yours to keep

Here's what I hadn't thought about. The people swinging hammers on these sites largely aren't local, because they can't be. There aren't 9,000 spare electricians sitting in a mid-size Texas city. So contractors fly in traveling trades from all over, cover housing and per diem, and pay very well for six to eighteen months of very long weeks.

It's a concert load-in at industrial scale. The trucks roll in, every hotel in town sells out, the show happens, the trucks roll out, and somebody stays behind to sweep the arena. Abilene had to approve new RV parks to have somewhere to put the people building the future.

While that was happening, average rent in town climbed roughly 50% to near $2,600 a month — reporting noted it briefly got cheaper to rent in Austin than in Abilene. A local shelter worker told reporters he'd started getting calls from data center workers who couldn't find anywhere to stay, and that the time it takes a family to land permanent housing had stretched from about a month to as long as six.

So the boom is real. It's just that a good chunk of the paychecks leave with the trucks, and the rent doesn't come back down on the same schedule. Studies of older energy boomtowns found construction turnover running well past 100% a year — those workers are already thinking about the next site before this one is finished.

The jobs that stay are a different trade entirely

This was my biggest wrong assumption walking in. I figured some slice of the build crew just moves inside when it opens. Mostly they don't, because it isn't the same job.

Building the thing is heavy construction — steel, conduit, switchgear, concrete, cranes. Running it is critical-facilities work: monitoring, cooling systems, power distribution, swapping hardware, responding at 3 a.m. when something trips. Different certifications, different rhythm, different hiring manager. A guy who wired the place has to go get retrained to work in it.

That path does exist, and it's better than I expected. A 12-week technician program in Red Oak, Texas placed 21 of its first 22 graduates into data center jobs at $22 to $40 an hour — including one man who'd been doing janitorial work at a data center and came out the other side as a critical infrastructure technician. Microsoft has run a similar community college pipeline in Virginia, Iowa and Washington for years.

The catch is arithmetic. That Red Oak class caps at twelve students. Twelve is a wonderful number if you're one of the twelve. It is not a plan for the several thousand people whose badges stopped working the week the campus went live.

So what does the county actually keep? And what I'd watch next

The research is less dramatic than either side's talking points. Brookings looked at counties that got their first large data center and found real growth in data processing and telecom work — but net job creation landing around 100 to 200 per county over a full decade. Coverage of that work noted local promises were regularly overstated by roughly threefold.

Georgia Tech researchers found something sharper: location decides everything. Metro counties saw employment, wages and new businesses tick up. Non-metro counties showed no measurable employment gain at all, while electricity prices in data center areas rose around 5%. Their line was that the wage and job growth promised during recruitment is unlikely to show up on its own.

And the price tag keeps surfacing. Virginia's data centers reported roughly 1,610 jobs in fiscal 2025 against about $1.9 billion in tax benefits. Georgia, Texas and Virginia are each now looking at north of a billion dollars a year in foregone revenue. More than a hundred communities have passed moratoriums and over 300 state bills were filed in the first weeks of 2026, which tells you the argument has moved out of the comment section and into statehouses.

What I still couldn't pin down: whether any of these deals get clawed back when the job counts come in short. It seems to depend entirely on how the individual agreement was written, and a lot of them don't appear to have been written tightly. If you find one that was, I'd genuinely like to see it.

This is the kind of rabbit hole Byte Bungalow goes down on camera — the turbines, the substations, the water, and the fine print nobody reads at the council meeting. Come watch the video version and subscribe on YouTube, because the next one is about who ends up paying for the power.

Common questions

Keep reading