a16z is circulating two charts built from U.S. government and bank research, framed under the headline "the AI buildout is printing electrician and HVAC jobs." The underlying claim: Goldman Sachs Global Investment Research, using data from the U.S. Bureau of Labor Statistics and Haver Analytics, attributes more than 300,000 construction jobs since 2022 to data-center-exposed trades, with roughly 75,000 of those added in the past year alone. A second, separate chart from the same account, sourced directly to BLS JOLTS data, plots data-center construction spending against construction job openings going back to mid-2024.
The correlation the headline implies is much newer than the chart's date range
The JOLTS chart plots exactly two years of data, July 2024 to July 2026, and the two lines it shows don't actually track each other for most of that span. From July 2024 through most of 2025, data-center construction spending climbs in a fairly steady line while overall construction job openings swing up and down without a clear matching trend — job openings actually bottom out around January 2025 while spending keeps climbing uninterrupted. It's only in the shaded window covering roughly the final few months of the chart — from around spring 2026 to the most recent data point — that the two series visibly turn upward together: spending accelerates from the high $50 billions to around $75 billion (SAAR), and job openings climb from roughly 230,000 to over 300,000 in the same stretch. The relationship the title asserts is real, but it's a recent inflection inside a two-year chart, confined to the last few months of it, not something the data has shown consistently the whole time.
Two ways to read "accelerating," and they don't point the same direction
The second chart's own headline calls this "accelerating," and that's true in one specific sense: the cumulative-employment panel shows the combined total for HVAC, electrical, utility, and commercial building construction climbing in a visibly steepening curve into 2026, consistent with roughly a quarter of the entire 2022–2026 gain landing in just the most recent year. But the same report's year-over-year growth-rate panel tells a more moderate story — the growth rates for these specific trades ran well above overall construction through 2022 and 2023, then spent 2024 and 2025 converging back down toward the broader construction rate, both ending up in the low single digits by 2026 rather than the wider gap seen earlier. Those aren't contradictory: a growth rate that's cooled while compounding on an already-larger base can still produce a bigger absolute number every year, which is exactly what the cumulative chart shows. But it means "acceleration" here describes the pace of new hires in raw numbers, not a re-acceleration in the underlying growth rate — worth keeping those two claims separate rather than treating one chart's headline as covering both panels.
The "roughly double" claim holds up
a16z's own text — "these trades are growing ~2% a year, roughly double construction overall" — is consistent with what the growth-rate panel shows for the most recent period: the data-center-exposed categories sit at roughly double the overall-construction line by 2026, even after two years of the gap narrowing from a wider one. That's a real, checkable claim against the chart as shown, not an exaggeration of it.
Solid data, told by an interested party
The underlying sources here are hard to dismiss: BLS JOLTS is the U.S. government's own official jobs-opening survey, and Goldman Sachs Global Investment Research is a bank publishing research under its own name, not an anonymous chart account. Neither is the kind of self-graded, unverifiable claim this blog treats skeptically elsewhere. What's worth naming is who's amplifying it and why: a16z is one of the largest venture investors in the AI infrastructure buildout this data describes, with a direct financial stake in the public and political read on that buildout staying favorable. "The AI buildout is printing electrician and HVAC jobs" is a genuinely useful counter-narrative to bubble and labor-displacement anxieties circulating elsewhere in AI discourse — and it happens to be the exact narrative a firm with money riding on continued AI capex has every reason to want told well.
Why the construction demand exists in the first place
This data measures an effect; this blog's own earlier coverage of the power-and-siting side of the AI buildout covers a cause worth connecting to it: interconnection queues running four to seven years in many U.S. regions, and industry estimates that roughly half of planned U.S. data-center projects may be delayed or cancelled in 2026 for power and permitting reasons rather than capital. Every one of those projects that does get built, delayed or not, still needs the same electricians, HVAC technicians, and commercial-construction crews this data is counting — which is also the labor-supply side of the GPU-and-chip commitments labs and cloud providers keep announcing: the hardware orders get the headlines, but building the buildings those chips sit in is its own, separately measurable economic story.
What to expect next
- Watch whether the spend-and-openings correlation holds up with more data. The co-movement the JOLTS chart highlights is only a few months old as shown; a longer run of data will show whether this is a durable relationship or a short-term coincidence.
- Watch whether the growth-rate convergence continues or reverses. If data-center-exposed trades keep converging toward overall construction's growth rate, the "roughly double" framing has a shelf life; if it re-widens, that's a stronger signal of a distinct AI-driven labor market forming inside construction.
- Watch for other analysts' occupational categorizations. Which BLS categories count as "data-center-exposed" is a judgment call Goldman made; a different cut of the same underlying data could show a different-sized effect.
- Watch this as a leading indicator if AI capex ever slows. Construction job openings move faster than most economic data; a genuine pullback in data-center spending should show up here before it shows up in chip-order announcements.