Texas oil fields and decommissioned energy infrastructure are being repurposed as sites for AI data center development, leveraging existing power and land assets.
Texas is rapidly becoming a datacenter mecca for the AI boom, driven by abundant natural gas supplies necessary to power multi-gigawatt-scale facilities. Meta announced its 29th datacenter project this week, which will be built in El Paso and eventually scale to a gigawatt of compute capacity by 2028. The El Paso facility will be powered by El Paso Electric, which sources power predominantly from natural gas, nuclear, and solar, with Meta stating that all energy consumed at the site will be matched with renewable energy. The initial investment is expected to cost around $1.5 billion and will be built in phases.
Simultaneously, the Wall Street Journal reports that Nvidia-backed AI developer Poolside is working with GPU rental outfit CoreWeave to deploy a two-gigawatt facility on West Texas oil fields. The companies plan to power the facility using onsite generators fueled by natural gas drawn directly from the source, arguing that proximity to these resources will cut costs and speed up development. The facility will be built in phases, with CoreWeave serving as the anchor customer for the first 250 megawatts of capacity beginning in 2027. While awaiting the full facility's completion, Poolside plans to rent a 40,000 GPU cluster of Nvidia GB300 rack systems from CoreWeave. The complete facility is expected to cost approximately $16 billion, not including compute hardware, though Poolside executives hope to reduce costs through a modular design and construction approach.
On the other side of Texas, ECL is developing a one-gigawatt datacenter in Houston that will be powered by hydrogen fuel cells. Lambda is set to be ECL's first tenant, though despite plans to bring the first 50 megawatts online this summer, as of late September the facility was not quite ready. Additionally, OpenAI's flagship Stargate campus is under construction in Abilene, where partners Crusoe and Oracle recently brought the first 200 megawatts online of what is expected to scale to 1.2 gigawatts by the end of next year. When complete, Oracle founder and CTO Larry Ellison announced that the site will house 450,000 Nvidia GPUs, equivalent to 6,250 Nvidia GB200 NVL72 rack systems.
These ambitious projects underscore the enormous capital requirements of AI infrastructure expansion. A single Nvidia NVL72 rack is estimated to cost $3.5 million, though large-scale deployers like Oracle, CoreWeave, and others are likely receiving volume discounts from Nvidia.