ECHELON INFRASTRUCTURE WORKFORCE REPORT · HUB ARTICLE · WEDNESDAY, AUGUST 19, 2026

Who will build America’s AI data centers? The labor problem

The Echelon Infrastructure Workforce Report is a Wednesday series on the licensed workforce building America’s data centers, power systems, and advanced infrastructure. Government records establish what is verified. Echelon analysis is identified as a read. Coverage limits remain visible.

The AI data center construction boom has a labor problem.

North America had more than 66 gigawatts of data center capacity under construction in August 2026, according to JLL. Seventy-seven percent of that capacity was being built outside the traditional core markets.^[7] The power has to be generated and delivered. The buildings have to be wired, cooled, piped, tested, commissioned, and maintained.

Software cannot do that work.

Electricians, plumbers, pipefitters, HVAC and refrigeration technicians, welders, equipment operators, fiber technicians, and licensed contractors turn capital commitments into operating infrastructure. Their availability will shape which AI projects open on schedule, which run over budget, and which wait behind other projects competing for the same people.

Echelon Reports is beginning a weekly examination of that workforce. We maintain and update public licensing records across the United States, preserve historical snapshots, and track changes in the licensed trades. This article is the hub for that series. Every Wednesday, we will publish a new finding about the people and companies building America’s data centers, power systems, advanced manufacturing plants, and other large infrastructure.

The short answer

Is there a skilled-trade labor shortage affecting AI data center construction?

Yes, but there is no honest national headcount that can tell a developer whether a specific project has enough labor. The shortage is local, trade-specific, and time-sensitive. A project may face adequate general construction labor while still lacking licensed electricians, controls technicians, pipefitters, or workers with critical-facility experience.

How many additional workers could the AI buildout require?

A September 2025 CSIS model estimated that generative AI infrastructure could require about 74,000 to more than 140,000 additional skilled-trade positions by 2030, depending on the buildout scenario. Its range included 19,000 to 37,000 electricians and 12,000 to 25,000 HVAC pipefitters.^[1]

What does Echelon currently measure?

Our August 17, 2026 working baseline contains 258,353 active, tiered electrical, plumbing, and HVAC credentials across the states and classifications included in the first release. That total includes 163,103 electrical credentials, 61,388 plumbing credentials, and 33,862 HVAC credentials.

These are credentials, not 258,353 unique workers. They do not prove that a worker is employed, available, willing to travel, or qualified for a data center assignment.

The numbers that define the problem

Measure	Current figure	What it tells us
North American data center capacity under construction	More than 66 GW	The construction wave is already in the field, not merely announced.[7]
Share under construction in frontier markets	77%	Much of the work is moving beyond established data center labor pools.[7]
Potential additional skilled-trade demand from generative AI by 2030	About 74,000 to more than 140,000	The range is large, but every scenario requires material workforce expansion.[1]
Potential additional electricians	19,000 to 37,000	Electrical labor is one of the clearest constraints in every AI buildout scenario.[1]
Potential additional HVAC pipefitters	12,000 to 25,000	Cooling and mechanical systems create a second major labor requirement.[1]
Annual electrician openings projected by BLS	81,000	The market must replace workers who retire or leave while also supporting new demand.[2]
Annual plumbing and pipefitting openings projected by BLS	44,000	AI projects compete with every other user of the same trade workforce.[3]
Annual HVAC openings projected by BLS	40,100	Cooling demand arrives in a labor market already handling replacement pressure.[4]
Additional construction workers ABC says the industry must attract in 2026	349,000	Data centers are competing inside a wider construction labor shortage.[6]
Echelon active tiered credentials in the initial covered database	258,353	This is the starting measurement base, subject to explicit coverage limits.

The numbers are large, but the national totals can mislead. A credential in Minnesota does not solve a project delay in West Texas. A residential electrician is not automatically ready for medium-voltage distribution, mission-critical commissioning, or a hyperscale construction schedule. An apprentice is part of future capacity, but supervision ratios and training requirements limit how quickly that person can substitute for a journeyman.

The useful question is not, “Are there enough electricians in America?”

The useful question is: Does this market have enough qualified labor for this project, during this construction window, after accounting for every other project drawing from the same workforce?

AI is becoming a physical construction industry

Public discussion of artificial intelligence still centers on models, chips, and electricity. That leaves out the work between a utility interconnection and a running server hall.

A large AI data center needs utility service, substations, switchgear, backup generation, power distribution, cooling plants, piping, controls, fire protection, racks, fiber, security systems, and commissioning. Each system brings its own labor requirements, licensing rules, and inspection sequence.

The scale of the electrical load explains why this matters. The Department of Energy reported that data centers consumed about 4.4 percent of U.S. electricity in 2023. DOE projected that share could reach 6.7 to 12 percent by 2028, with total consumption rising from 176 terawatt-hours in 2023 to between 325 and 580 terawatt-hours in 2028.^[5]

That demand reaches far beyond the data center fence. New generation, transmission, distribution, substations, and equipment manufacturing also require skilled labor. The electrician who might wire an AI campus may instead be working on a semiconductor fab, hospital expansion, battery plant, transmission upgrade, or utility project.

McKinsey estimated that procurement and installation spending for data center mechanical and electrical systems could exceed $250 billion by 2030.^[12] Money can order equipment. It cannot compress apprenticeship hours, licensing requirements, or field experience into a shorter calendar.

Why the national workforce statistics do not answer the local question

The Bureau of Labor Statistics provides the best broad view of U.S. occupations. It counted 818,700 electrician jobs in 2024 and projected 9 percent growth from 2024 to 2034. BLS also projected 81,000 electrician openings a year, with many caused by retirement, occupational transfers, and other exits from the labor force.^[2]

For plumbers, pipefitters, and steamfitters, BLS projected 44,000 annual openings and 4 percent employment growth.^[3] For HVAC and refrigeration mechanics and installers, it projected 40,100 annual openings and 8 percent growth.^[4]

Those figures establish the national pressure. They cannot tell us how many people can work on a particular project.

Five gaps separate a national occupational estimate from actual project capacity:

1. Geography. Construction labor markets are local until travel becomes economical. Per diem, overtime, housing, licensing, and schedule length determine whether workers move.
2. Licensing. States classify electricians, plumbers, HVAC workers, contractors, and qualifiers differently. Reciprocity ranges from useful to nonexistent.
3. Skill. A license establishes a legal credential. It does not establish medium-voltage, controls, clean-room, industrial refrigeration, liquid-cooling, or commissioning experience.
4. Availability. An employed worker is not spare capacity. A license database does not reveal current assignments, overtime, backlog, or willingness to travel.
5. Timing. Two viable projects can become a labor bottleneck when their peak electrical or mechanical schedules overlap.

This is why claims about a single national “worker gap” need care. A shortage is not one pile of empty jobs. It is a series of trade, market, and schedule collisions.

What Echelon’s license database sees

Government licensing records offer a different view of the workforce. They identify active credentials, classification, status, location, business affiliation in some states, and the dates when records appear or change.

Echelon’s first covered baseline contains:

Trade	Apprentice	Journeyman	Master	Total active credentials
Electrical	47,743	55,789	59,571	163,103
Plumbing	25,467	21,173	14,748	61,388
HVAC	14,853	16,310	2,699	33,862
Total	88,063	93,272	77,018	258,353

The initial electrical worker-tier coverage includes Colorado, Idaho, Kentucky, Minnesota, Oklahoma, Texas, Virginia, and Wisconsin. Plumbing coverage includes Colorado, Idaho, Kentucky, Minnesota, Oklahoma, Rhode Island, and Wisconsin. HVAC coverage currently includes Idaho, Kentucky, and Oklahoma.

California and Ohio are present in Echelon’s database, but their current tables measure contractor licensing rather than apprentice, journeyman, and master workforce tiers. California’s August 15 snapshot contains 27,940 clear contractor licenses carrying the C-10 electrical classification, 17,381 carrying C-36 plumbing, and 11,470 carrying C-20 HVAC. Ohio contains 3,504 active electrical contractor licenses, 2,740 plumbing, 2,641 HVAC, 866 hydronics, and 693 refrigeration licenses. Adding those records to the worker-tier table would mix contractor credentials with worker credentials and inflate the apparent workforce. We will measure them separately as contractor and qualifier capacity. California worker-level coverage will require a separate ingestion of the Division of Labor Standards Enforcement electrician certification and trainee records. Ohio does not publish an equivalent statewide apprentice-to-master roster through its construction licensing board.

The underlying source observations range from July 31 through August 16, 2026. We aggregated the baseline on August 17.

The imbalance in state coverage matters. It would be easy to put 258,353 in a headline and call it a workforce count. That would be wrong.

Our database measures active licenses and registrations inside defined state and classification coverage. One person may hold credentials in more than one state or at more than one level. Some states license workers. Others emphasize contractors or qualifying individuals. Some regulate HVAC statewide, while others leave important parts of the trade to local authorities.

We will publish those limitations with the data because the alternative is false precision.

What license records can tell us

A license record is not a worker-availability record. It is still useful.

Repeated snapshots can show:

• New apprentice registrations
• New journeyman and master credentials
• Renewals, lapses, and reinstatements
• Contractor entry and exit
• Qualifier movement between businesses
• Geographic expansion into another state
• Concentration of scarce qualifying credentials
• The time required for a new cohort to move through the licensing structure

That movement is the beginning of a capacity picture. If a market adds apprentices but not journeymen, its pipeline may be growing while near-term delivery capacity remains tight. If master credentials are concentrated among a small number of contractors, the market may be more fragile than its total license count suggests. If project starts accelerate while active credentials remain flat, wage and schedule pressure should follow.

None of those signals confirms that a shortage caused a delayed project. They identify where to look before the delay becomes public.

The apprentice pipeline cannot solve a 2027 deadline by itself

Apprenticeships matter because licensed trades cannot expand without new entrants. They also take time.

The CSIS high-case scenario estimated a need for roughly 37,000 additional electricians and 25,000 HVAC pipefitters by 2030. Its recommended response included a 50 percent expansion of registered apprenticeships in priority trades.^[1]

Google has begun funding that pipeline. In 2025, the company announced support for an electrical training effort intended to reach 100,000 electrical workers and train 30,000 new apprentices. The program aims to increase the electrical workforce pipeline by 70 percent over five years.^[8] In June 2026, Google expanded its skilled-trades commitment to $50 million, with a stated goal of helping train more than 300,000 workers through unions and trade associations.^[9]

Meta and CBRE launched a separate multiyear program in April 2026 to recruit and train thousands of fiber technicians for U.S. data center construction.^[10] The federal AI Action Plan also called for initiatives to increase high-demand occupations such as electricians and HVAC technicians while accelerating data center and semiconductor-fab development.^[11]

These programs are evidence that major buyers see labor formation as part of infrastructure strategy. They are not evidence that the shortage has been solved.

An apprentice entering a program in 2026 does not become a journey-level worker when a 2027 project schedule needs one. Training capacity, instructor availability, completion rates, required work hours, and supervision rules determine how much of the pipeline converts into deployable capacity.

Near-term projects will still rely on the existing workforce, travel labor, overtime, prefabrication, schedule sequencing, and competition among contractors.

The bottleneck will move

The first constraint in many AI markets is power. Once developers secure power, the constraint can shift to equipment. Once the transformers and switchgear are available, the project still needs enough qualified people to install and commission them.

Labor shortages rarely stop every project at once. They change the order of completion.

Large owners can pay travel packages, lock in contractors early, standardize designs, and secure manufacturing capacity. Their best response is to reserve labor before the market tightens further. That response makes labor less available to smaller projects, municipal work, hospitals, manufacturers, and local contractors.

Contractors will respond by recruiting across state lines, acquiring firms with licenses and customer relationships, expanding apprenticeship sponsorship, using prefabricated assemblies, and protecting key qualifiers. Some will grow. Others will accept more work than their supervision and field capacity can support.

The pressure will not stay inside data centers. It will move through every market that shares the same electricians, pipefitters, HVAC technicians, and equipment.

What this Wednesday series will measure

The Echelon Infrastructure Workforce Report will publish every Wednesday.

We will separate slow structural capacity from weekly movement. That distinction matters because government rosters, apprenticeship counts, BLS employment data, and project announcements update on different schedules.

The series will examine five questions:

1. Is the licensed workforce growing?

We will track active credentials by trade and level, then distinguish apprentice formation from journey-level and master capacity.

2. Where is demand arriving?

We will map data centers, power infrastructure, semiconductor plants, advanced manufacturing, and other projects that compete for the same workers.

3. When will projects collide?

A market can absorb one large project and still fail under three overlapping schedules. Timing matters as much as announced capital.

4. Which credentials and contractors are hard to replace?

Qualifier concentration, specialty classifications, reciprocity barriers, and contractor control can turn a modest headcount issue into a serious execution risk.

5. What changed this week?

New licenses, lapses, reinstatements, contractor entry, qualifier movement, project milestones, and schedule changes will form the weekly pulse.

The work will lead to a formal Echelon capacity index after we have enough history to establish a defensible baseline, test the weighting, and separate real movement from changes in coverage. We will not declare the first observation “normal” simply because it came first.

What buyers should watch now

Developers, contractors, utilities, investors, economic-development groups, and equipment suppliers should watch the same early indicators:

• Apprentice growth without matching journeyman growth
• Flat active-license counts during accelerating project starts
• Multiple projects entering electrical or mechanical peak at the same time
• Rising dependence on a small number of master or qualifier licenses
• Interstate expansion by large contractors
• Increased use of travel labor, overtime, and per diem
• Project announcements in markets without an established industrial-trades base
• Training announcements that begin after the construction schedule already needs workers

No single signal proves a market is constrained. Several moving in the same direction deserve attention.

The finding to carry forward

America’s AI race is also a construction race.

The country is trying to add tens of gigawatts of data center capacity while replacing workers who retire, expanding the grid, building factories, and maintaining everything already in service. The same licensed trades sit underneath all of it.

Echelon’s initial database does not claim to count every skilled worker in America. It establishes something more useful: a defined, auditable starting point that can be updated, compared, and improved.

Every Wednesday, we will show what changed.

Workforce intelligence in this cluster

This page is the demand-side hub. The connected reports separate licensed supply, worker-level credential capacity, project demand, and verified mobility.

---

Methodology and limitations

Echelon compiled the license baseline from state government records in its covered electrical, plumbing, and HVAC classifications. We counted records that met each source’s active-status rules and normalized them into apprentice, journeyman, and master tiers where the source supported that mapping.

The unit is an active credential or registration. It is not necessarily a unique person, employed worker, available worker, contractor business, or worker qualified for critical-facility construction.

State systems use incompatible classifications and publication schedules. Coverage is incomplete and uneven by trade. Counts may include cross-state and cross-tier duplication. We will preserve source-native classifications, snapshot dates, and coverage disclosures as the system expands.

External demand estimates describe scenarios or broad occupational markets. They should not be interpreted as verified vacancies for data center work. Project-level capacity analysis requires location, schedule, trade intensity, project probability, contractor commitments, and the ability of workers to travel or obtain reciprocal credentials.

Frequently asked questions

Is AI causing an electrician shortage?

AI data center construction is adding electrical demand to a workforce already serving utilities, manufacturing, commercial construction, housing, and maintenance. It is a material source of pressure, but it is not the only cause of electrician scarcity.

How many electricians are needed for AI data centers?

CSIS estimated that generative AI infrastructure could require about 19,000 to 37,000 additional electricians by 2030, depending on the adoption and construction scenario.^[1] That is a modeled range, not a current vacancy count.

Which skilled trades are needed to build a data center?

Data center construction uses electricians, plumbers, pipefitters, HVAC and refrigeration technicians, welders, fiber technicians, equipment operators, controls specialists, fire-protection workers, and other construction trades. The mix changes with the design, cooling system, power architecture, and project phase.

Why are electricians important to AI infrastructure?

AI data centers require large, reliable electrical systems. Electricians install and maintain distribution equipment, feeders, backup systems, controls, and the connections that carry power from utility service to computing equipment.

Can apprentices fill the data center labor gap?

Apprentices are essential to long-term workforce growth, but they cannot immediately replace experienced journey-level workers. Training hours, supervision ratios, licensing rules, and project complexity limit how fast apprentice enrollment becomes independent field capacity.

Does an active license equal an available worker?

No. An active license proves that a credential met the issuing authority’s active-status rules when the record was collected. It does not prove employment, availability, specialty experience, location, or willingness to travel.

What will Echelon publish every Wednesday?

The Echelon Infrastructure Workforce Report will track licensing changes, workforce formation, contractor and qualifier movement, project demand, geographic pressure, and evidence of emerging trade bottlenecks.

Sources

1. The Generative AI Infrastructure Challenge
2. Electricians: Occupational Outlook Handbook
3. Plumbers, Pipefitters, and Steamfitters: Occupational Outlook Handbook
4. HVAC Mechanics and Installers: Occupational Outlook Handbook
5. DOE Evaluates Increase in Electricity Demand from Data Centers
6. Construction Industry Must Attract 349,000 Workers in 2026
7. North America Data Center Report Midyear 2026
8. Google Electrical Worker and Apprentice Training
9. Google Skilled Trades Training Investment
10. Meta and CBRE LevelUp Technician Training
11. America’s AI Action Plan
12. AI Power: Expanding Data Center Capacity
13. PORTS-Piketon: The contractor and skilled-trades map behind America’s largest proposed AI campus

MIKE WARNER | ECHELON REPORTS
mike@echelonreports.com | linkedin.com/in/jmichaelwarner | x.com/_Mike_Warner

Public-records intelligence. Not accounting, legal, or investment advice.