America's artificial-intelligence buildout is acquiring a new geography. The obvious centres of gravity remain Northern Virginia, Texas, Arizona and a handful of large cloud regions, but the next wave of investment is increasingly constrained by something much less portable than software talent: electricity.
In American Commerce Review's 2026 audience survey, 39% of 14,726 respondents described affordable electricity as a very significant factor in future AI and data-centre investment, while another 35% called it critical. Only 9% regarded it as slight or insignificant. That is an unusually decisive result for a cost item that, until recently, was often treated as one line in a wider site-selection model.
Power demand is moving faster than conventional infrastructure planning
The external data point in the same direction. Berkeley Lab's latest US data-centre electricity work estimates that data centres could consume 9.5% to 15.3% of US electricity by 2030, with a central estimate of 11.8%. The earlier Department of Energy-backed analysis put 2023 consumption at 176 terawatt-hours, or 4.4% of US electricity use. The scale of the change matters because electricity systems are not built on software release cycles.
New generation can sometimes be added quickly. High-voltage transmission, substations and large-load interconnections usually cannot. The Department of Energy has separately highlighted grid-capacity constraints and interconnection timelines as barriers to data-centre development, with individual proposed sites now requesting power measured in gigawatts. A state can therefore have cheap land and generous tax treatment yet still lose a project if the connection date falls outside the developer's investment window.
The winning states will sell certainty, not just low prices
That changes what an attractive AI-infrastructure market looks like. The cheapest wholesale power is not automatically the best proposition if a developer cannot secure firm capacity, redundancy and a credible energisation date. Conversely, a state with somewhat higher power costs can remain competitive if utilities and regulators can make capacity available sooner and with fewer planning surprises.
This is why American Commerce Review's state-level AI coverage increasingly treats electricity as part of the investment stack rather than an adjacent energy story. Virginia's fibre density and cloud ecosystem still matter. Texas benefits from scale, energy development and a comparatively flexible power market. Other states can enter the contest where industrial sites, generation and transmission capacity line up.
A data-centre boom is not automatically a local economic boom
There is an important qualification. Large data-centre investments create construction demand, tax revenue and infrastructure spending, but their permanent employment intensity is very different from a semiconductor fab or advanced manufacturing plant. The local economic case depends on the tax structure, who pays for network upgrades, water requirements and whether new generation serves the wider system as well as the data-centre campus.
For policymakers, the useful question is therefore not how many billions of dollars have been announced. It is how much capacity has been contracted, when it will be energised, what network reinforcement is required and how the cost is allocated.
What the survey does and does not show
The ACR survey is an audience survey, not a nationally representative poll of US businesses. Its value is in identifying the priorities of a large, commercially engaged readership and then testing those priorities against independent evidence. On electricity, the alignment is unusually strong: readers say power availability is decisive just as federal and laboratory data show data-centre load becoming materially larger.
The next phase of the AI infrastructure race will therefore be easier to read through megawatts, interconnection dates and grid investment than through headline project values alone.
| Response | Share |
|---|---|
| Critical | 35% |
| Very significant | 39% |
| Moderately significant | 17% |
| Slightly significant | 6% |
| Not significant | 3% |
Frequently asked questions
Why is electricity becoming a constraint on US AI data centres?
AI data centres require very large, reliable power connections, while generation, transmission and substations can take years to permit and build. The constraint is increasingly the date at which firm capacity can actually be delivered.
Is the ACR survey nationally representative?
No. It is an American Commerce Review audience survey of 14,726 respondents. National claims in this article are benchmarked separately against US government and Berkeley Lab evidence.