The Geography of AI: Why Virginia and Texas Keep Winning Data Center Projects

Policy & Markets By EnergyInnoCore Editor · Published Oct 20, 2026 · Updated Oct 21, 2026 · 11 min read

You don't have to work in the data center industry to notice the pattern: data centers are heavily concentrated in a few places, above all Virginia and Texas. Why? I've discussed this with friends who have spent years in data center development and in project finance, and one thing becomes clear quickly: these projects do not land where they do by accident.

For a long time, data center location was driven by familiar factors: reliable electricity, good fiber connections, reasonable land costs and a local government that would not make development unnecessarily difficult. Those factors still matter, but AI has changed the scale. A conventional data center was discussed in tens of megawatts; today developers routinely plan campuses of hundreds of megawatts, and some at gigawatt scale. At that size, choosing a site looks less like commercial real estate and more like locating a refinery, power plant or major industrial complex. Northern Virginia was already well positioned when this transition began. Texas entered the AI era with a different set of strengths. The two arrived at roughly the same destination by different paths, and both now face the limits of their own success.

Virginia Had the Network Before AI Arrived

To understand Virginia you have to start with Northern Virginia, and specifically Ashburn. Northern Virginia is still regarded as the largest data center market in the world. A 2024 study by Virginia's Joint Legislative Audit and Review Commission (JLARC) found that the region accounted for roughly 13% of reported global operational data center capacity and about 25% of capacity in the Americas. The study credits a combination of strong fiber infrastructure, reliable power, available land, proximity to major customers and tax incentives.

That foundation was built over decades. Northern Virginia became one of the main crossroads of the commercial internet, and telecom carriers, network providers, cloud companies, government agencies and eventually hyperscalers built there because so much traffic already passed through the region. Once that network density existed, a powerful feedback loop began: data centers attracted more networks, and more networks attracted more data centers. Customers wanted to be close to one another because physical distance still matters in computing; every additional mile of fiber adds latency, even if it is measured in fractions of a millisecond. For ordinary web traffic that may not matter much, but for cloud services, financial transactions, AI inference and machine-to-machine communication it can matter a great deal.

The result was something like gravitational pull. Once Northern Virginia had accumulated enough fiber routes, internet exchanges, cloud regions and data centers, it became very difficult for other markets to reproduce the same ecosystem, which is why Ashburn became known as "Data Center Alley." For developers, building near that ecosystem removes a major uncertainty: they are not wondering whether fiber will eventually arrive, because it is already there.

Power made the fiber valuable

Fiber alone does not run servers; electricity does. Northern Virginia benefited from a large, reliable utility system operated mainly by Dominion Energy, backed by the broader PJM transmission network. Data center operators do not simply want cheap electricity. They want power that is available around the clock, deliverable at industrial scale and supplied through a highly reliable grid. Attractive prices mean little if the grid cannot physically deliver the megawatts. For years, Northern Virginia offered a favorable combination of availability, reliability, infrastructure and cost, and the market grew rapidly.

Eventually, the market began running into the consequences of that growth. Large data center demand is now putting heavy pressure on transmission, substations, generation and utility planning. That creates an important distinction: Virginia became dominant partly because power was available, and its challenge today is whether power availability can keep pace with the next generation of AI demand. This is a recurring pattern in infrastructure: a successful market attracts investment, investment creates demand, and demand eventually consumes the spare capacity that attracted investment in the first place. Northern Virginia may be the clearest example of that cycle. (We describe the resulting grid pressures in We Can Serve You, But Not Yet.)

Texas Came from a Different Direction

Texas does not have the same decades-long network concentration as Northern Virginia. What it has is scale: large amounts of land, abundant energy production, big metropolitan areas, major fiber routes, a business-friendly development culture and, perhaps most important for the AI era, the habit of thinking about infrastructure at very large scale.

Dallas-Fort Worth is already one of the largest data center markets in the country. CBRE reports that DFW has become North America's third-largest colocation market, with inventory growing more than 40% year over year and several hundred megawatts under construction, nearly all of it preleased. But Texas is no longer just a Dallas story. Development interest is spreading into Central Texas, West Texas and other regions, largely because AI developers are looking for something increasingly hard to find elsewhere: room. Projects such as Crusoe's AI campus planned to grow from 200 MW to 1.2 GW in West Texas show the scale now being contemplated.

Land matters more than people realize

A modern AI campus needs much more than a building full of servers. It may need one or more electrical substations, backup generation, battery systems, cooling equipment, water infrastructure, transmission rights-of-way, large setbacks, security zones and, increasingly, space for on-site power generation. Once campuses reach 500 MW or 1 GW, the land requirement becomes substantial, and inexpensive land near Ashburn is no longer one of Northern Virginia's advantages.

In parts of Texas, developers can assemble hundreds or even thousands of acres. That matters because developers increasingly value optionality. They may build the first 200 MW today but want room to expand to 500 MW later. They may connect to the grid initially but keep land available to add natural gas generation, fuel cells, batteries, solar or other behind-the-meter infrastructure. They may want multiple fiber entrances, separate substations and redundancy between buildings. Land provides those options, and options have value. In project development, the cheapest land is not always the best value; the more important question is what the land allows you to do later.

Then there is power

Texas is one of the world's major energy-producing regions, with enormous natural gas production, one of the largest wind fleets in the country, rapidly growing solar, existing thermal generation and a large independent grid managed by ERCOT. Competitive electricity markets have long made Texas attractive to large industrial loads, and data centers fit naturally into that picture.

AI has changed the math here too. ERCOT is handling an extraordinary volume of large-load interconnection requests. In June 2026, Texas regulators approved a "Batch Zero" process that studies large users of 75 MW and above together rather than one at a time, so ERCOT can understand how much new load the system can support, where, and with what transmission upgrades. By August 2026 the situation was even more striking: Governor Greg Abbott said ERCOT was considering more than 474 GW of connection requests, more than five times the grid's record peak demand, and ordered a comprehensive audit before data center projects could continue through the interconnection process. Clearly not all 474 GW will be built, since queues always include speculative and duplicate projects. Still, the figure shows that developers see Texas as one of the places where very large AI infrastructure might still be possible, and that perception alone has created a development rush.

Texas knows energy development

A harder-to-quantify factor is experience. Texas has spent generations developing oil fields, gas infrastructure, pipelines, power plants, transmission lines, wind and solar farms and industrial facilities. The result is an ecosystem of engineers, developers, land professionals, contractors, power marketers, financiers and regulators accustomed to major infrastructure projects. AI data centers increasingly resemble energy projects as much as technology projects; when a campus needs 1 GW, power is no longer just a utility connection but one of the central development workstreams.

That makes Texas attractive even as grid capacity tightens. A developer might conclude: "I can't get all 500 MW from the grid immediately, but perhaps I can secure 200 MW, add on-site gas generation, contract renewable energy, install storage and expand in phases." That kind of thinking fits naturally into the Texas energy market. (See our discussion of this approach in When the Power Won't Show Up, Do You Build Your Own?.)

Regulation and Incentives Matter Too

Developers usually talk about power and land first, but regulatory conditions can quietly decide whether a project succeeds. A developer may spend hundreds of millions of dollars before the first server is installed, and every extra year spent on rezoning, permitting, utility studies, litigation or local approvals adds carrying costs and risk. Both Virginia and Texas historically created relatively favorable environments. Virginia's tax incentives helped establish Northern Virginia as a technology hub. Texas offers sales tax exemptions for qualifying data centers covering equipment, electricity, cooling systems, network hardware and other infrastructure, with larger projects required to meet substantial investment, job and power-contract thresholds. Because data centers are so capital-intensive, with servers, transformers, switchgear, cooling and network hardware all adding up, removing sales tax from even part of that spending can materially change project economics.

The regulatory welcome is changing

Both states are now asking a new question: how much data center development is too much? Residents are increasingly concerned about electricity costs, water use, transmission lines, noise, land use and the relatively small number of permanent jobs some facilities create. Virginia has introduced measures to manage the industry's impact, including a temporary electricity consumption tax on data centers that took effect in July 2026. Texas is moving in a similar direction: Governor Abbott has directed regulators to ensure data centers pay for the electrical infrastructure built to serve them rather than passing costs to residential customers, and has called for water-efficient cooling, better reporting of power and water use, and stronger protections for nearby communities.

That does not mean either state is turning against data centers. It means the relationship is changing. Early in a development cycle, communities ask how to attract these projects; after hundreds arrive, the question becomes how to manage them. Virginia is well into that second phase, and Texas is getting there quickly. (For one federal angle on cost allocation, see our analysis of the Ratepayer Protection Act.)

Fiber Still Decides More Than People Think

Power gets most of the attention today, understandably, but fiber remains critical. An AI data center cannot operate as an isolated island. Training clusters need enormous internal bandwidth, cloud regions need multiple redundant external connections, and inference facilities increasingly need to sit close to users, enterprise customers and cloud networks. Northern Virginia's advantage here is hard to overstate after decades of telecom development. Dallas-Fort Worth also sits at the intersection of major national fiber routes, and CBRE notes that developers are moving farther south along the I-35 corridor partly because connectivity there is improving. Data center markets can expand outward: once power, fiber and transportation extend from an established hub, developers can move to cheaper land while staying connected to the original ecosystem. That happened around Ashburn, and a similar pattern is emerging in Texas.

In the End, Developers Are Buying Certainty

People often expect one simple answer to why so many data centers are in Virginia and Texas: cheap electricity, tax incentives or available land. Site selection rarely works that way. Developers are buying a combination of enough power, multiple fiber routes, land that allows expansion, an understandable regulatory environment, access to contractors and financing, transmission, water and a reasonable expectation that the facility will actually be operating in a few years. The most valuable commodity may be certainty. A theoretically perfect site is worthless if the utility cannot say when power will arrive; cheap land is not cheap if transmission upgrades take six years; a tax incentive does not help if local opposition blocks construction; and abundant power is of limited use without reliable fiber.

Virginia succeeded because many of these pieces came together early. Texas is succeeding because it offers many of the same ingredients at a scale increasingly attractive to AI developers. But neither state has unlimited capacity. Virginia has the world's deepest existing data center ecosystem, but power availability, land constraints and community resistance are becoming harder issues. Texas has land, energy resources and a culture built around large infrastructure, but its grid is now confronting hundreds of gigawatts of proposed demand. The concentration of data centers in these two states tells us where yesterday's advantages were. The more interesting question now is how much more Virginia and Texas can realistically absorb before the next generation of AI infrastructure starts looking somewhere else.

EnergyInnoCore Editor

Independent analysis of power, grid and policy for AI data centers. Every article lists its sources. About the editor · Editorial policy

This article is for general information only and is not engineering, legal, financial or investment advice. See the disclaimer.