Where America Is Building AI Data Centers—and How They Will Get Power

A new AI data center appears on a map.

What does that dot actually tell you?

It may mark an operating facility.

Or a planned campus that still needs permits.

It may represent 20 MW.

Or a multi-gigawatt project built in phases.

And even a fully financed project still needs one more thing:

a credible path to power.

A map tells you where activity is concentrated. A project’s status and power plan tell you whether that activity can become useful compute.

By the end of this article, you will be able to read a U.S. data-center map or project announcement, separate directory counts from electrical capacity, identify how power is supposed to arrive, and spot the questions that still need answers.

Map showing where U.S. data centers are concentrated and how major AI campuses plan to get power

The Map Is a Snapshot, Not a Power Meter

The original July 21, 2026 map used the state listing totals displayed by Data Center Map on that date.

Virginia, Texas, and California stood out in the directory snapshot.

Choropleth map of U.S. data center directory listings by state in July 2026

Snapshot date: July 21, 2026. Directory listings show concentration of activity, not audited AI capacity or state electricity demand.

This warning is essential.

A directory listing is not a standardized unit.

one small facility may count once
one gigawatt-scale campus may also count once
one campus can appear as multiple buildings or phases
planned and operating projects may sit in the same directory ecosystem

So the map answers:

Where is data-center activity concentrated?

It does not directly answer:

How many megawatts are operating in each state?

The Contexta Map Integrity Test

Before turning any data-center map into a conclusion, ask four questions.

What is being counted?
Listings, buildings, campuses, MW, or investment dollars?
What status is included?
Operating, under construction, planned, land-banked, or all of them?
What date is the snapshot?
Live directories change.
Does one unit have a standard size?
If not, count and capacity are different measures.

This prevents one of the easiest mistakes in AI-infrastructure reporting: treating a facility count like a power-capacity chart.

The Geographic Shift: New Projects Are Moving Rural

Pew Research Center analyzed Data Center Map records current to February 19, 2026.

It found more than 3,000 operating U.S. data centers and more than 1,500 in development.^[1]

The pattern is more interesting than the total.

67% of planned projects were in rural areas
87% of operating facilities were in urban areas
39% of planned projects were in counties with no current data center
75% of planned projects were in the South and Midwest

That is a major geographic shift.

The older data-center economy clustered around established urban connectivity hubs.

The AI buildout is placing more value on a different package:

land + high-voltage access + room for substations + expansion space + a credible power schedule.

Why Rural Land Is Becoming More Attractive

One recurring public question is:

Why build on farmland or open rural land instead of old malls, factories, or office parks?

There is no single answer.

But large AI campuses can need hundreds of acres, dedicated substations, transmission access, large cooling systems, construction staging areas, backup generation, and space for later phases.

Practitioner discussions repeatedly point to speed-to-power, proximity to high-voltage lines, lower land cost, and easier expansion as key reasons rural sites can become attractive.

That does not mean every rural location is suitable.

A cheap field with no fiber, no transmission capacity, no water plan, and no workforce is not automatically a data-center site.

The Contexta Site Triangle

A useful first screen is:

Land → Connectivity → Power

For older cloud and network hubs, connectivity could dominate location decisions.

For very large AI campuses, the balance is changing.

The power side of the triangle can now decide whether a site moves from announcement to operation.

Six U.S. Projects Show Six Different Power Stories

The table below is not a ranking or a complete U.S. construction list.

It is a set of current examples showing that “AI data center” does not imply one standard power model.

Project	Status / timing	Public scale	Power plan
Abilene, Texas Crusoe / Oracle / Microsoft	Existing buildings operating; additional Microsoft campus under development	About 2.1 GW projected across the combined Abilene footprint	Grid-connected infrastructure plus a new 900 MW campus with dedicated onsite power and batteries
Childress, Texas Crusoe / Lancium	Construction expected to begin Q3 2026	1.0 GW grid-connected campus	Grid interconnection and energy orchestration managed with storage support
Hyperion, Louisiana Meta	Under construction and expanding	5 GW of compute capacity, using Meta's terminology	Utility agreement funds seven gas plants, three grid batteries, nuclear uprates, and purchased power
AWS Mississippi campuses Amazon	Multi-campus buildout under construction / expansion	$25B statewide planned investment; no single comparable MW figure stated in the cited source	Amazon says it is funding $300M of grid improvements and supporting 616 MW of new renewable projects
Port Washington, Wisconsin Oracle / partners	Construction progressing; customer delivery expected in 2H27	Four buildings on 500 acres; electrical campus MW not stated on the public project page	Oracle says 70% zero-emission energy target and ~2,000 MW of new wind, solar, and battery resources added to Wisconsin's grid
Saline Township, Michigan Oracle / OpenAI / Related Digital	Three buildings under construction; customer delivery begins in 2H27	Three large buildings; comparable electrical MW not publicly stated in the cited project material	DTE resources plus project-funded battery storage, transmission, and onsite substation infrastructure

These examples show why public “scale” numbers require care.

A company may report:

IT capacity
facility power
generation capacity
compute capacity
investment dollars
building area

Those units should not be placed in one ranking as if they measure the same thing.

The Contexta Project Status Ladder

Another common mistake is treating every announcement as operating capacity.

Use this ladder instead:

Announced → Permitted → Under Construction → Energized → Operating → Expanding

A project can move forward on buildings while the grid path is still being completed.

So “under construction” does not automatically mean “power-ready.”

Five Power Strategies Are Emerging

1. Regional grid connection

This remains the basic model.

But a “grid connection” can imply new transmission, substations, transformers, studies, and years of utility work.

2. New utility generation

Very large campuses can become large enough to shape utility resource planning.

Meta’s Louisiana project is a clear example: its current agreement funds gas generation, batteries, nuclear uprates, and purchased power.^[4]

3. Behind-the-meter or onsite generation

Developers are increasingly looking at generation that can be deployed with the campus rather than waiting entirely for the normal grid path.

Reuters reported on September 29, 2026 that U.S. data-center developers are showing strong interest in smaller gas turbines because they can sometimes be deployed faster than large grid or generation projects.^[8]

This can improve time-to-power, but it may raise lifetime energy cost, fuel, emissions, and local permitting questions.

4. Batteries and flexible load

Batteries do not create electricity.

They can shift it through time, bridge interruptions, support power quality, and help manage peaks.

Some large customers can also agree to curtail load during stressed grid conditions.

5. Add resources to the wider grid

Wind, solar, batteries, nuclear uprates, or other resources do not have to sit beside the server hall.

They can change the regional supply mix while the data center remains grid-connected.

The Power Plan Is Becoming Part of the Site Plan

The old mental model was:

choose land → build data center → buy electricity.

The emerging model is closer to:

Choose land + secure grid path + shape supply plan + build compute together

This is one reason the new AI map can look different from the older cloud-data-center map.

Why Communities Are Asking Harder Questions

The geographic shift toward rural areas is happening at the same time public concern is increasing.

Pew’s August 2026 survey found 60% of U.S. adults were not too or not at all comfortable with a new data center operating in their area.^[2]

Among rural adults, the share saying data centers are mostly bad for:

the environment rose to 50%
home energy costs rose to 46%
nearby quality of life rose to 45%

Those views do not prove that every project will cause those outcomes.

They do tell us which questions communities want answered.

The Contexta Local Consequence Check

When a new campus is proposed, residents and local officials can ask:

Power: How many MW are requested now and at full buildout?
Grid: What transmission, substation, or transformer work is required?
Cost: Who pays if the grid upgrade is underused or the project changes?
Generation: Is new onsite or utility generation required?
Water and cooling: What cooling architecture is proposed?
Noise and air: What backup or onsite generation will operate, and how often?
Jobs and tax: Which benefits are construction-phase, and which remain after opening?

This is more useful than asking whether data centers are simply “good” or “bad.”

It turns a broad local debate into project-specific questions.

How to Read the Next AI Data Center Announcement

Use this eight-point decoder.

Status: announced, permitted, under construction, energized, operating, or expanding?
Scale unit: IT MW, facility MW, generation MW, compute capacity, acreage, or dollars?
Initial phase: what opens first?
Final buildout: what is the long-term campus target?
Grid connection: approved, studied, funded, or still proposed?
Power equipment: what substations, transformers, transmission, batteries, or generators are needed?
Cost allocation: who pays for the infrastructure and carries downside risk?
Community footprint: what changes in land, water, noise, traffic, tax base, and jobs?

That is enough to separate many real project milestones from a large headline number.

The Contexta Map-to-Power Decoder

The entire article can be reduced to four layers:

Map → Status → Power Plan → Local Consequence

Map tells us where activity is clustering.

Status tells us how real and how advanced the project is.

Power Plan tells us whether the site has a credible path to operation.

Local Consequence tells us who gains, who pays, and what infrastructure changes around the project.

What Should You Watch Next?

Rural shift: does the South/Midwest share keep rising?
Energization: how many announced multi-GW projects actually receive power on schedule?
Behind-the-meter generation: bridge solution or permanent operating model?
Rate design: how do utilities keep large-load costs from shifting to other customers?
Community response: do permitting standards tighten as public concern rises?
Power-unit discipline: will companies publish more comparable IT/facility MW figures?

The Main Idea

America’s AI data-center map is changing.

Established hubs still matter.

But the next wave is moving toward places that can combine land, expansion room, connectivity, and a credible path to large blocks of electricity.

The map shows where the buildout is moving. The status tells you how real it is. The power plan tells you whether it can turn on.

That gives us the next question:

When a large project needs new grid infrastructure, who should pay for it?

Continue Reading

Cheap Power Is Not Always Cheap — see why connection time changes site economics.
What Must Be Built to Power the AI Data Center Boom? — follow the electricity-to-compute infrastructure chain.
Why Power, Not Chips, May Limit the AI Data Center Boom — understand the grid bottleneck behind the map.

Key Terms

directory listing: an entry in an industry database; it is not a standardized unit of power capacity
operating: a facility that is in service rather than merely announced or under construction
energized: connected to sufficient electrical service to begin operating the intended load
planned project: a project in development that may still require permits, construction, financing, or grid milestones
grid interconnection: the technical and regulatory process for connecting a large new load to the power system
behind-the-meter generation: generation serving a site on the customer side of the normal utility meter or connection structure
curtailment: temporarily reducing load when grid conditions require it
cost allocation: deciding which parties pay for new generation, transmission, substations, and other infrastructure
full buildout: the planned final scale of a multi-phase campus
site selection: choosing a location based on land, connectivity, power, cooling, risk, cost, and expansion needs

Sources

Status checked September 30, 2026. The July 21 map remains a dated directory snapshot and should not be read as an audited capacity map. The Map Integrity Test, Site Triangle, Project Status Ladder, Local Consequence Check, and Map-to-Power Decoder are The Contexta analytical frameworks. Project scale terms follow the cited sources and are not assumed to be directly comparable across companies.