Texas is racing to become a capital of the AI boom, as cities from Abilene to Austin court massive data centers that promise jobs and tax revenue. Grid planners and environmental experts warn that this same rush could overload the Electric Reliability Council of Texas system and deepen the risk of blackouts just as extreme heat and winter storms grow more punishing. The stakes are national, because the United States is on track for AI data centers to consume a far larger share of electricity in only a few years.
A collision is coming between the fast growth of digital services and the slower work of building power plants, power lines, and water systems. The question is not whether Texas will host more data centers, but whether the state can keep the lights on for everyone else while it does.
AI demand meets a fragile grid
Across the country, analysts expect data centers to account for 12 percent of national electricity consumption by 2028, according to a recent brief from a project at the Belfer Center for Science and International Affairs that focused on AI data centers. That forecast helps explain why Texas, with its large fleet of gas plants and fast-growing solar and wind output, has become such a magnet for server farms. Yet the same grid is still remembered for deadly outages during severe winter storms, and it operates as an island, with limited connections to other states that could share power in an emergency.
That isolation raises the stakes when new, always-on loads arrive. Unlike a factory that can sometimes scale back output, AI data centers are designed to run around the clock, with limited flexibility once contracts are signed. If planners misjudge how much supply is needed, the shortfall lands on households and small businesses that have no say in where these facilities are built or how they are powered.
Texas becomes a data center magnet
Developers have zeroed in on Texas as the next big hub for AI infrastructure, and a recent report predicts that the state will soon be the top market for data centers, with facilities getting larger over time. That same research highlights companies such as Bloom Energy, a California-based firm that provides onsite power generation for these electricity-hungry campuses, as part of the new ecosystem forming around this buildout.
On the ground, the scale of individual projects is staggering. In October, plans for the Stargate I campus in Abilene called for a 1.2 gigawatt power capacity, which equates to roughly 1,200 megawatts, once the site is fully built out. The same reporting projected that data centers across Texas could require roughly 100,000 MW by 2030, a figure that dwarfs many current planning baselines and signals how much strain this sector could place on the grid if even a portion of those projects move forward at speed, as described in analysis of large.
Speculative projects and planning whiplash
The surge of interest is not limited to a few headline campuses. By December, so many data center requests had flooded into Texas that energy experts said the demand was impossible to meet, with more speculative projects on paper than the state can realistically build. According to reporting on this wave of speculative development, interconnection queues are now packed with proposals that may never see shovels in the ground.
From a planning point of view, that glut is a problem in its own right. Grid operators must study each request, model the load, and decide what new lines or substations are needed, even if the project is just a placeholder. The result is a kind of whiplash, with engineers racing to plan for hypothetical demand while real-world upgrades lag behind, and communities left guessing which projects will actually appear near their homes and farms.
Large loads outpacing wires and plants
Texas is already feeling the strain of this mismatch between digital ambition and physical infrastructure. In October, grid planners described a “rapid explosion” of large load users looking to connect to the state’s electric system, and those facilities are being built faster than traditional transmission infrastructure can keep up. At the same time, the system had 85.5 gigawatts of capacity in August 2023, a figure that shows how quickly new demand from AI campuses could eat into existing headroom if multiple projects come online at once, according to reporting on grid planning struggles.
The key tension is timing. Transmission lines can take most of a decade from concept to completion, once landowner disputes, environmental reviews, and financing are factored in. By contrast, a well-funded data center developer can move from announcement to operation in just a few years. Unless Texas finds ways to speed up grid upgrades or slow down approvals for the largest loads, the wires that carry power may become the weakest link in the AI economy.
Water, land, and the ‘giant soda straw’ effect
Power is not the only resource at stake. A study by the Houston Advanced Research Center and the University of Houston examined how data centers affect groundwater, and one local described their impact as “like a giant soda straw.” Researchers found that heavy withdrawals for cooling caused a water level drop of 16 feet in a year, an extreme shift that shows how a single industrial user can change local hydrology, according to the HARC and University work.
That “giant soda straw” metaphor applies just as well to electricity. In a rural county that hosts a big AI campus, the data center can draw more power than the surrounding towns combined, turning the facility into a straw stuck straight into the grid. Residents see the land and water footprint up close, while the digital services that drive this consumption, from language models to image generators, may feel abstract and far away.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
