Who Builds the Compute: An AI Workforce Data Book

This data book covers four parts of the AI workforce that builds and runs large-scale computing: constructing and commissioning data centers, operating them (data center operations, run by data center operators and technicians), producing the chips inside them, and writing the software that runs on top. Each group is identified from job-title keywords. That makes the groups easy to reproduce, but titles are an imperfect guide, and the limits are listed in the method section.

Job postings show what employers are trying to fill. They do not show how many people were hired. This section compares the number of US postings each group had in September 2024, September 2025 and September 2026, to show hiring trends for 2026 against earlier levels.

The headline measure is the change in monthly postings between two single months, with each year shown separately. Total US postings in the data were lower in September 2026 than in September 2024, so each group’s share of all US postings, shown as a secondary view in the table, rises by more than its raw count. For the data center groups, most of the growth came in the first year. Read together, the figures are an early look at future workforce trends for AI infrastructure.

Electrical and power, cooling and mechanical, and construction roles all grew far faster than compute software roles. Semiconductor and technician roles grew the least. Because the comparison uses single months, a few postings can move the figures, so read the line chart as well as its end points.

Two views of experience are shown here. The first is the minimum experience employers ask for in job postings. The second is the experience of US professionals who currently hold these roles, taken from Draup’s profile data.

For data center technicians the two views do not line up, which points to an experience gap inside the wider AI skills gap. 84% of technician postings with a stated requirement ask for three years or less, while 12% of the 12,997 US technician profiles in the data show three years or less. The early-career group is a small part of the technician profiles in the data.

Postings are counted by US metro area and split into two kinds of titles: data center titles, which cover the physical build and operation, and compute software titles. The two kinds cluster in different places. Data center roles make up 96% of the postings in Columbus and 94% in Reno, against 46% in the San Francisco Bay Area, where software titles lead.

Each metro below is flagged for data center construction, a sign of AI infrastructure investment, and shows how much its data center postings grew. Of the metros shown, 15 have data centers under construction, 5 have projects planned or proposed, and 2 had no under-construction project in the sources checked. Growth is the change in a metro’s monthly data center postings, excluding technician roles, from September 2024 to September 2026. Small metros start from small counts, so their percentages swing widely. Hover a row to see the evidence behind its flag.

Draup profile data shows where people who now hold data center roles worked before, a view of AI talent trends in the data center workforce. This section covers people who started a data center role in 2022 or later and whose previous job was in a different metro. Results are shown as indexes, not as headcounts. The main Talent Flow Index compares each metro’s net arrivals, meaning people arriving minus people leaving, with the size of its data center workforce, so a small metro with a large gain scores higher than a big metro with a small one. The largest result is set to plus 100. A second column, the volume index, ignores size and compares net arrivals in people with the largest net gain.

Metros with construction under way mostly gain data center talent: 8 of the 11 that have enough data are net gainers, against 1 of 3 where projects are planned or proposed and 0 of 2 where no under-construction project was identified. The two lenses differ. Washington DC has the largest net gain in people (volume index +100) but a modest gain for its size (+9), while Abilene’s gain is large for its small base (+100), which a small base can exaggerate. The pattern is the opposite of the one in posting growth, where metros with construction were not the fastest growing. With few metros in each group, it shows direction, not proof.

The war for talent in data centers is not fought by the same kinds of employers in every role group. 5 of the ten largest employers of electrical, cooling and mechanical roles are engineering, construction, real estate services or equipment companies. For compute software, none of the ten largest employers fall in those categories (0); the list is mostly chipmakers and large technology companies, plus a job board.

Demand for the AI workforce is strongest in the trades that build and power data centers. Electrical, cooling and construction roles grew far faster than compute software, employers accept early-career technicians while the visible workforce skews experienced, and talent is moving toward the metros where data centers are being built. For any organization planning around AI infrastructure investment, the open questions are which roles it will need and which tasks AI will change first.

Postings, experience, locations and employers. These come from Draup’s job-posting table in the data warehouse, limited to postings located in the United States, for September 2024 through September 2026, queried on October 16, 2026. Job titles are matched on the normalized title field. Counts are postings, not openings or hires, and can include reposts. Growth is the change in monthly postings between September 2024 and September 2026, with September 2025 as the midpoint, and shares of all US postings are a secondary view.

Workforce supply and experience. These come from Draup’s professional profile snapshot, filtered to profiles in the United States that carry the named role, queried on October 13, 2026. A profile can carry more than one role.

Construction flags. Flags come from Draup news signals in the Center Setup and Expansion category, captured from January to October 2026 and searched by metro. CBRE’s North America Data Center Trends H1 2026 report supplies megawatts under construction for its primary markets. Where Draup signals had no coverage, as in Des Moines and Omaha, a public project tracker (Baxtel, Epoch AI) is used and named in the table. Each flag rests on news reporting and has not been checked against permits.

Metro demand growth. For each metro, monthly data center postings in September 2026 against September 2024, excluding technician titles because bulk postings distort them, from the same warehouse table and US filter as the other demand figures. Metros with small counts, such as Abilene, show very large percentages.

Talent flow. The flows come from Draup’s professional profiles. A profile counts when its current position is a data center role (Data Center Technician, Engineer, Operations Manager, Analyst or Director of Data Center Operations) that started in 2022 or later. The origin is the metro of the previous position, which must have ended no more than a year before the new one began. Only corridors with at least five movers are kept, to limit noise. Corridor values are scaled to the largest corridor. Each metro’s net arrivals are divided by the number of data center profiles currently in that metro, and the results are scaled so the largest absolute value is 100. The volume index uses net arrivals alone, scaled the same way. Internal transfers are included.

Limits. Postings are limited to those located in the United States. Titles are a proxy for the work. Electricians hired directly by contractors may not carry “data center” in the title, so electrical roles are probably undercounted in the postings. Overall posting volume swings from month to month, which is why growth is shown as a share of all postings. Metro growth rates for smaller metros rest on counts in the tens to low hundreds a month and move easily. The talent flow index rests on profile data, which under-represents skilled trades, metros without a corridor of enough movers are not scored, and a small workforce base, as in Abilene, can exaggerate the size-adjusted index. Hiring difficulty and talent supply by metro are not included in this edition.

© 2026 Draup. All rights reserved.