Texas is seeing a rapid expansion of data centers, a trend that directly impacts local communities and the state’s power grid. Prime Data Centers operates two adjacent facilities south of Dallas Love Field with a combined capacity of 20 megawatts—enough electricity for roughly 15,000 homes. Across the Stemmons Freeway, CleanArc Data Centers is developing a campus slated for 245 megawatts, entering the realm of hyperscale operations.
Even larger projects on the horizon
Tech giants are planning even bigger sites. Microsoft is building a 2‑gigawatt facility in Reeves County, while Meta plans a 1‑gigawatt center in El Paso. At the extreme end, the Fermi Project Matador proposal envisions up to 17 gigawatts near Amarillo—about half the total capacity that Oncor supplies to Dallas, North Texas and surrounding areas.
Why the power surge matters
According to a U.S. Congressional report, roughly half of a data center’s electricity use powers the servers themselves; the remainder fuels cooling systems that keep the hardware from overheating. Unlike a stadium that only spikes power on game days, data centers require a constant, high‑level supply.
In August 2025, Oncor, the regional utility with a peak capacity of 33 gigawatts, reported a queue of 200 gigawatts of interconnection requests, 186 gigawatts of which came from data centers. By the first quarter of this year, that queue had grown to 271 gigawatts, underscoring the accelerating demand.
Grid stability concerns
University of Texas at Austin associate professor Hao Zhu warns that the rapid, large‑scale fluctuations typical of data‑center loads can destabilize the grid. “Data centers have a lot of power fluctuations, and this is a destabilizer to the power grid because they have very fast cycles,” Zhu said.
These fluctuations add complexity to an already independent Texas grid, which many residents view with skepticism regarding reliability.
AI’s appetite for energy
Artificial‑intelligence workloads drive much of the growth. MIT Technology Review measured the energy needed for various open‑source AI models. A modest 8‑billion‑parameter Llama model consumes about 57 joules per text query, while a 400‑billion‑parameter version uses roughly 3,400 joules. When accounting for cooling and supporting hardware, total energy use roughly doubles.
Video generation is especially power‑hungry; a five‑second clip created with an open‑source model required about 3.4 million joules—approximately one kilowatt‑hour, the same amount billed for a typical household appliance running for an hour.
Local implications
For communities hosting these facilities, the surge in power demand can affect rates, reliability, and future development plans. Residents and local officials are urged to stay informed about grid capacity and to engage with utilities as new projects seek interconnection.
Original reporting: Texarkana Gazette — read the source article.
