Bain: $6.5 Trillion AI Buildout Hinges on Power and Electrical Labor

Bain & Company projects $5 trillion to $6.5 trillion in cumulative data center spending through 2030, enough to add nearly 150 gigawatts of compute capacity, and says the AI market would need to approach $6 trillion in annual revenue by 2031 to justify it. In an interview published October 8, a Bain partner told Construction Dive that the immediate limits for builders are power and electrical labor, and that not every announced project will make sense to build.

What Happened

Bain's 2026 Technology Report, released September 29, includes a chapter on whether the industry can physically deliver the data center capacity that AI demand implies. Its base case calls for $5 trillion to $6.5 trillion of spending to add nearly 150 gigawatts of compute capacity by 2030, almost tripling the global total, with most of the growth in the United States, according to Bain & Company.

On October 8, Construction Dive published a question and answer interview with Peter Hanbury, a Bain partner and one of the report's authors, focused on what the forecast means for contractors. Hanbury called the spending level "conceivable" given how fast the largest technology companies have raised their capital budgets, but he questioned whether AI companies will generate enough revenue to sustain construction at that pace.

The Numbers

• Annual spending: AI infrastructure spending could reach about $1.5 trillion a year by 2031, per Construction Dive.
• Revenue required: Hanbury said capital expenditure runs at roughly 25% of industry revenue, so $1.5 trillion in annual spending implies an AI market of roughly $6 trillion, per Construction Dive.
• Current outlays: Microsoft, Google, Amazon, Meta and Oracle could spend about $780 billion on capital expenditures in 2026, nearly five times what they spent three years earlier. Not all of that goes to data centers, per Construction Dive.
• Installed capacity: Global data center IT power rises from 79 gigawatts in 2025 to 228 gigawatts in 2030 in the base case, with a low case of 188 gigawatts and a high case of 262 gigawatts, per Bain & Company.
• Electricity demand: Global data center electricity use grows from 388 terawatt-hours in 2023 to 1,228 terawatt-hours in 2030, per Bain & Company.
• Campus size: Five years ago a 50 megawatt facility was considered large. Hyperscalers now plan campuses of 5 gigawatts or more that cost $150 billion to $200 billion each, per Bain & Company.
• Cooling: Revenue from cooling technology grows from $7.9 billion to $28.4 billion over the same 2023 to 2030 period, excluding China, per Bain & Company.

The report compares the scale of U.S. data center investment, as a share of the economy, with past spending on railroads, telecommunications networks and the electrical grid, per Bain & Company.

Four Constraints at Once

The report's central argument is that power, chips, skilled labor and permitting are all constrained at the same time, so the shortage will not resolve on its own and site-by-site workarounds will not be enough, according to Bain & Company.

Power. Grid connection is delaying and freezing projects, and adding grid capacity takes four years or more. Transformers and other critical equipment are scarce. On-site, behind-the-meter generation eases the bottleneck but adds regulatory exposure, longer equipment lead times and project complexity, per Bain & Company. Hanbury made the same point to Construction Dive: a data center can be built in a few years, but major new grid capacity can take four years or longer.

Labor. The report identifies shortages of electrical workers, mechanical tradespeople and cooling specialists, and says meeting demand will require recruiting, training and retaining talent at rates well above historical levels, per Bain & Company. Hanbury named electrical labor as the biggest trade bottleneck, especially high-voltage, substation and mission-critical electrical talent. He expects mechanical and pipefitting trades to come under pressure as liquid cooling scales, along with controls and commissioning specialists, per Construction Dive.

Equipment. Graphics processors, memory, optical networking and other components are being reserved years ahead, and operators without long-term supply agreements are effectively shut out of the queue, per Bain & Company. Because lead times are long, decisions made today determine capacity several years out.

Permitting and local opposition. Permitting runs through local, state and national levels and adds time and uncertainty even when every other input is secured. In the first quarter of 2026, local opposition blocked or delayed at least 75 projects worth $130 billion, nearly matching the $156 billion disrupted in all of 2025, per Bain & Company. Opposition centers on water use, energy prices and noise.

The report notes that in July 2026 New York became the first U.S. state to pause construction of the largest data centers, and that 14 other states have introduced similar legislation, per Bain & Company. Our earlier coverage of the New York data center pause has more on that policy.

How Long It Takes to Build

A 1 gigawatt data center typically takes one to four years to build, even on a site with existing power, and longer if new generation or transmission is required, according to Bain & Company. The report cites build times ranging from about one year to 3.6 years across recent large projects. The fastest example it describes involved repurposing an existing factory and leasing dozens of gas turbines, an approach that drew a lawsuit over air permits.

What Bain Says Has to Change

Hanbury told Construction Dive that four conditions must hold for the investment level to be achievable:

• AI has to move beyond efficiency and productivity uses into new sources of revenue, such as autonomous systems, physical AI and new products.
• Physical bottlenecks in power, chips, skilled labor and permits have to ease. On power, he called for more generation, faster interconnection, behind-the-meter capacity, storage and better coordination between utilities and technology companies.
• Capital models have to share more risk among developers, infrastructure investors, utilities and governments, including jointly funded power infrastructure.
• The industry has to be more disciplined about which projects get built.

The report itself lays out system-level responses, per Bain & Company. Providers may turn to shared power procurement, colocation and jointly funded grid upgrades. In the near term, behind-the-meter generation, mostly natural gas, is among the few ways to obtain large blocks of power on a data center schedule. Renewables, storage and virtual power plants are growing, while new large-scale nuclear capacity is unlikely to arrive until the mid-2030s. The report also expects communities and governments to require binding agreements, and it cites a Lancaster, Pennsylvania, contract that caps water use and sets clean energy, noise, emissions and local hiring requirements for a data center project.

Implications for Owners, Developers, Contractors and Subs

Contractors should underwrite the pipeline, not just the project. Hanbury gave Construction Dive four questions to ask of any data center opportunity: Is the power real? Is the customer and financing commitment real? Is the permission to build real? Is the design stable enough to build? He added that permission now includes community support on power use, water, noise and emissions. Those questions are a useful screen for any contractor or subcontractor deciding how much capacity to commit to a single client or program, because a stalled megaproject can strand crews, bonding capacity and prefabrication investment.

Delivery is shifting from projects to programs. Hanbury said the industry has to move toward industrial-scale programs with integrated power and site planning, earlier procurement, more prefabrication, modular designs and coordinated portfolios of contractors and suppliers, per Construction Dive. He noted that the slowest constrained input sets the schedule for the whole program. For trade contractors this favors firms that have invested in off-site fabrication and can commit to long-term partnerships across multiple sites.

Scope is expanding beyond the building. Contractors are increasingly working on substations, transmission, interconnection and sometimes on-site generation and storage as part of the same program, per Construction Dive. General contractors without utility-side experience will need partners who have it.

Design is less stable than it looks. Hanbury described "chip-to-grid" codesign, where changes in processors drive rack density, which in turn changes networking, cooling, electrical architecture and the power source, per Construction Dive. That means late design changes are a structural feature of this work, and contracts should allocate that risk clearly.

Owners outside the data center sector are affected too. Manufacturers, healthcare systems, distribution operators and institutions compete with these programs for the same electricians, pipefitters, switchgear and transformers. With equipment being reserved years ahead and electrical labor identified as the tightest trade, owners of conventional commercial and industrial projects should lock in electrical equipment early, bring key trade partners on during design and carry realistic lead times in their schedules. Owners planning any large new load should also start the utility conversation early, given the four-year horizon for new grid capacity.

Community terms are becoming part of the deal. With $130 billion of projects disrupted in a single quarter, developers of any large industrial facility should expect to address water, noise, energy cost and local hiring before approvals, not after.

What to Watch

• Whether hyperscaler capital spending keeps rising in 2027 or begins to level off.
• More state-level pauses or siting rules following New York's.
• Transformer, switchgear and electrical labor availability on non-data-center projects.
• Growth in behind-the-meter gas generation and jointly funded grid upgrades tied to specific campuses.
• Cancellations or delays among announced gigawatt-scale campuses, which would test Hanbury's point about project discipline.

Bottom Line

The forecast is very large and Bain considers it achievable, but only if revenue materializes and the physical constraints ease together. For builders the binding limits are power and electrical labor. For everyone else in the construction market, the practical effect is sustained competition for the same trades and equipment, which makes early procurement and careful vetting of project fundamentals more important than usual.

Image credit: Bain & Company. Image from the source article.

Sources

• Construction Dive, "A $6 trillion question hangs over the AI construction boom": https://www.constructiondive.com/news/a-question-hangs-over-ai-construction-boom/832470/
• Bain & Company, "AI Data Center Boom: Can We Build It If They Come?": https://www.bain.com/insights/ai-data-center-boom-can-we-build-it-if-they-come-technology-report-2026/