Where Are Data Centers Being Built, and Why Power Decides It

Ask where are data centers being built and most answers give you a list of states. The more useful question is why those states. Land, fiber and tax breaks matter, but in 2026 the deciding factor is usually electricity: whether a site can get enough of it, and when.

This guide follows the power data: Northern Virginia, Texas, the rest of the PJM region and Alberta, then the factors that decide a location. Figures come from EIA, PJM, ERCOT, the AESO and the IEA, last checked October 28, 2026. They change often, so check the date on each one.

Where are new data centers being built in the US?

Where are new data centers being built? Mostly where they already are. The IEA's Energy and AI report (April 2025) found nearly half of US data center capacity sits in five regional clusters, and that 50% of data centers under development were in those existing large clusters.

The grid data points to two regions above all:

• The PJM region, the grid that runs from Virginia and Maryland through Pennsylvania, New Jersey and Ohio to northern Illinois. Northern Virginia is its center of gravity.
• Texas, on the ERCOT grid, where requests to connect large loads have outrun anything the system has served before.

EIA's May 2026 analysis shows the effect. Between 2019 and 2025, commercial electricity sales in Virginia rose by nearly 30.0 million MWh, more than in any state except Texas. EIA puts Virginia's growth down mainly to data centers. In its September 2026 outlook, EIA named data center development as a main reason demand is growing fastest in the West South Central region, which includes Texas.

The concentration matters at state level too. The IEA reported that data centers already used more than 10% of the electricity supply in six US states in 2024, with Virginia highest at about 25%.

Elsewhere, projects go wherever a large block of power, land and fiber comes together, and more of them plan for on-site generation.

Why does Northern Virginia still lead?

Northern Virginia leads because the cluster is already there: fiber, existing campuses, and contractors and staff who know the work. EIA describes the PJM zone that covers most of Virginia as serving the largest concentration of data centers in the world.

The cost of that success shows up in the peak load numbers EIA reported in May 2026:

• Summer peak load in the zone was 23,905 MW in 2025, 23% above 2019.
• Winter peak in the 2025-26 season was 25,413 MW, 45% above 2019-20.
• PJM's 2026 forecast expects this zone to see the largest absolute rise in summer peak of any PJM zone from 2026 to 2030, growing an average of 5.4% a year over the next ten years. That is down from 6.3% in its 2025 forecast.

For a new tenant, this means capacity in the region exists but is competed for. New grid connections depend on transmission and substation upgrades, and the data center interconnection queue sets the date more often than the building does. The region stays the default for workloads that must sit near users, but it is a hard place to get a large new block of power quickly.

Why are so many data centers being built in Texas?

Texas draws builders with land, a single-state grid run by one operator, and abundant natural gas.

As of mid-2026, ERCOT said large users had requested more than 438,000 MW of new demand, and nearly 90% of those requests came from data centers. For scale, ERCOT's all-time peak demand is 85,508 MW, set on August 10, 2023. ERCOT itself says not all requests become projects.

To cope, ERCOT replaced one-at-a-time studies with a batch process approved in June 2026. The first group, Batch Zero, started in July 2026, with a deadline for financial commitments planned for June 2027. ERCOT also created two optional routes in:

• Bring your own generation. Large users that build on-site generation, such as gas turbines, can offset the transmission capacity they need.
• Controllable load. Users that let ERCOT cut their power during local grid stress can use part of the grid before the full build-out is done.

Gas supply is the other draw. EIA forecasts Permian Basin gas production at 29.2 Bcf/d in 2026 (as of August 2026), and more than 66% of the new US pipeline capacity planned for 2026 and 2027 starts in Texas (as of May 2026). That makes Texas a natural home for gas-powered data centers.

Which other US regions are getting new data centers?

Beyond Virginia, growth is spreading across the rest of the PJM region. In its 2026 load forecast (posted January 2026), PJM adjusted 14 of its zones for growth in data center load. Those zones cover parts of Ohio, Pennsylvania, northern Illinois, New Jersey, Maryland, West Virginia and Washington, DC, as well as Virginia.

PJM forecasts its summer peak reaching 222,106 MW in 2036, an increase of 65,733 MW over ten years, or 3.6% a year on average. Data centers are a main reason for that rise.

Note how PJM counts. Near-term large loads now need firm commitments, such as signed construction agreements, to count in full; longer-term projects are discounted as non-firm. That lowered its near-term forecast compared with the year before.

Other regions appear in the news, but without grid data as clear as this. An announced project is not a power connection.

Are data centers being built in Canada?

Yes, and Alberta is where public grid data shows the clearest demand. Alberta has a deregulated power market, natural gas, and cold weather that helps with cooling.

In June 2025, the Alberta Electric System Operator (AESO) said 29 proposed large-load projects, totaling more than 16 GW, were asking to connect. It set an interim limit of 1,200 MW for new large loads of 75 MW or more through 2028, the amount it said the grid could serve without hurting reliability.

The next phase leans toward sites that supply their own power. The AESO is designing a bring-your-own-generation process for large loads that self-supply on site or add new generation. As of its September 8, 2026 update, the process was still being finalized, with intake expected in fall 2026.

As in the US, where the grid is tight, projects that bring their own firm power move up the line.

Why does power decide where data centers go?

Power decides because it is the scarcest input and the slowest to arrive. An AI campus can need hundreds of megawatts, and few grids have that much spare at one point. The factors, roughly in order of how hard they are to fix:

1. Grid headroom. Is there spare capacity at a nearby substation, or does the site need new transmission? New lines take years.
2. Connection process and queue time. How the utility and grid operator study large loads, in what order, and how long it takes. Batch studies, firm-commitment rules and connection limits all change the answer.
3. Fuel for on-site power. If the grid is slow, can the site run on its own generation? That needs firm gas from a pipeline or gathering system, plus air permits. This is the case for behind-the-meter data centers.
4. Fiber. Training tolerates distance from users; real-time inference needs better connectivity.
5. Land and zoning. Large parcels with local support. Rural land is cheaper but needs roads, staff and services.
6. Tax and policy. Many states offer sales-tax exemptions on data center equipment. Incentives pick between workable sites; they rarely rescue an unworkable one.
7. Water and climate. Cooling design depends on both, and local water rules can limit some designs.

The first three are about power, which is why builders increasingly follow the power rather than the people. For a full checklist, see our guide to data center site selection.

How should you read data center location headlines?

Read every location number by asking what it counts. Most headline figures count requests or announcements, not buildings with power.

• Requests are not projects. ERCOT's 438,000 MW of large-load requests is about five times its record peak. The same developer may file for several sites.
• Forecasts get revised. PJM's 2026 forecast came in lower than its 2025 forecast for the near term, partly because it now discounts loads without firm commitments.
• Limits move. Alberta's 1,200 MW was an interim figure, and the long-term framework was still being written as of September 2026.

Our honest take: the public data is clear on direction and weak on detail. Virginia and Texas lead, the wider PJM region is growing, and grid operators everywhere are rationing new large-load connections. Exactly how many megawatts get built where, and when, is uncertain. Anyone who quotes you a precise figure for 2030 is quoting a forecast.