AI can move faster than the infrastructure required to support it
Artificial intelligence is usually described as a race for better chips, larger models and more computing power. But the next phase of the race may be decided by something far less glamorous: whether the physical world can deliver enough electricity, transmission, cooling, fuel, fiber and construction capacity to keep the machines running.
That changes the question.
The question is no longer simply, “Who can build the best AI?” It is becoming, “Where can the entire support system be assembled, connected and operated at scale?”
The demand is arriving faster than the system
The U.S. Department of Energy reported that data centers used roughly 4.4% of American electricity in 2023. Its published range for 2028 is approximately 6.7% to 12%. In energy terms, the report estimates an increase from 176 terawatt-hours in 2023 to between 325 and 580 terawatt-hours by 2028.
Those figures do not guarantee that every proposed data center will be built. They show the size of the demand now pressing against the system.
The International Energy Agency makes the underlying dependency plain: there is no AI without electricity for data centers. Yet generating electricity is only one part of the problem. Power must also be transmitted to the right place, stepped down through substations, delivered through transformers and switchgear, backed by fuel or storage, and converted into useful computing without exceeding local cooling and water limits.
A region can possess abundant energy in theory and still lack usable capacity in practice.
The bottleneck is a chain, not a single shortage
Public discussion often treats the problem as though one more power plant will solve it. The actual buildout depends on several systems arriving together:
- Generation and dependable fuel supply
- High-voltage transmission and substations
- Transformers, switchgear and other electrical equipment
- Cooling systems and, in some designs, significant water access
- Fiber routes and low-latency connectivity
- Skilled labor, financing, permits and community acceptance
Delay one of those layers and the value of the others can remain stranded.
This is why the most important signal may not be a national shortage. It may be a mismatch between where demand wants to go and where the supporting infrastructure can actually be qualified and activated.
Pressure does not disappear. It migrates.
When a preferred location cannot connect new load quickly enough, developers have several choices. They can wait, pay more, reduce the project, add on-site generation or search for another location.
That search creates a second wave of pressure.
Regions with attractive power, land, climate or tax conditions may begin receiving projects that could not move efficiently elsewhere. But the receiving region then has to prove that its advantage is real. Cheap land is not enough if transmission is weak. Cool weather is not enough if fiber is limited. Available generation is not enough if equipment, permits or financing cannot arrive on schedule.
The pressure can therefore move from a visible constraint—such as electricity—to a quieter one such as transformers, skilled trades, political permission or project finance.
This is the distinction Atlas is designed to watch: not merely where demand begins, but where it is forced to travel and which dependency becomes binding next.
What this means for communities
For communities positioned to receive new infrastructure, the opportunity can be substantial. Large projects can expand the tax base, create construction work, improve transmission and fiber, and attract supporting businesses.
The costs can also be unevenly distributed. Residents may face competition for power, water, land and public attention before the promised benefits are broadly felt. Local governments can be pressured to approve incentives or infrastructure commitments before the full burden is visible.
The central public-policy question is not whether data centers are good or bad. It is whether the agreement protects the host community while the project is still dependent on local approval.
That is when communities have the strongest opportunity to negotiate measurable protections: who pays for grid upgrades, how water use is managed, what happens if the project is delayed, which jobs are genuinely local, and how households are protected from cost shifting.
What Atlas is watching now
The current watch is not a prediction of collapse, shortage or investment performance. It is a structural observation.
AI demand is colliding with infrastructure that was not designed to expand at software speed. If the first-choice locations cannot assemble the full support stack, projects will be delayed, resized or redirected. The receiving locations will then be tested by their own transmission, equipment, connectivity, financing and political constraints.
The decisive map may not be the map of the cheapest electricity.
It may be the map of places where electricity, wiring, cooling, fiber, labor, finance and public consent can all arrive at the same time.
Research boundary
This briefing presents a dependency and pressure-migration assessment. It is not an investment recommendation, a claim of exact timing, or evidence that any specific project or region will succeed or fail.
Pressure is not price. Dependency is not profitability. Outcomes will be updated as evidence matures.
Official sources
1. U.S. Department of Energy
Source note: DOE summarizes the Lawrence Berkeley National Laboratory estimate that U.S. data-center electricity use could rise from about 4.4% of national consumption in 2023 to approximately 6.7%–12% by 2028.
2. Lawrence Berkeley National Laboratory
2024 United States Data Center Energy Usage Report
Source note: The underlying technical report estimates historical and projected data-center electricity demand and explains the assumptions and uncertainty behind the published range.
3. International Energy Agency
Energy and AI — April 10, 2025
Source note: The IEA examines electricity demand from AI and data centers, the supply required to serve it, and the implications for energy security, affordability and system planning.