The Texas AI Boom is Outpacing Water Regulations

The world’s largest artificial intelligence data center complex is being built in Amarillo, and it’s almost seven times the size of Central Park. 

On June 26, 2025, Fermi America, an AI development firm cofounded by former Texas Governor Rick Perry, announced that it will partner with the Texas Tech University System to transform 5,800 acres outside of Amarillo into 18 million square feet of data centers, four 1-gigawatt nuclear power reactors, and a dedicated natural gas plant. 

The site is colloquially known as Project Matador, though a building request submitted to the Nuclear Regulatory Commission lists its official name as the President Donald J. Trump Advanced Energy and Intelligence Campus. On July 10, two weeks after Project Matador’s announcement, the White House released the United States’ AI Action Plan, a 28-page strategy document calling for streamlined environmental permitting under federal law. Cornerstone environmental protections are being treated as mere hurdles in the race for AI dominance, and Texas’ water (and power) supply is struggling to keep pace. 

Texas is rushing to build massive artificial intelligence infrastructure without a planning system capable of assessing industrial water needs, experts warn. There are currently over 400 data centers operating or under construction in the Lone Star State. Potable freshwater is necessary at each campus for industry-standard evaporative cooling, a process in which water absorbs heat and either becomes too mineralized for reuse or is lost as vapor. Additional water is needed to cool power-generating systems.

 A small-to-mid-sized data center is estimated to require about 300,000 gallons of municipal water per day, while mega-campuses like Project Matador and OpenAI’s Project Stargate One in Abilene could draw millions. (Fermi America did not respond to requests for comment about its projected water use.) All of this is happening as Texas simultaneously acknowledges water scarcity: last year, legislators passed and voters approved a multibillion-dollar water program that will roll out over the next 20 years.

In late January, the Houston Advanced Research Center (HARC), a nonprofit that does research on energy demand and water scarcity, released a report projecting that the existing data centers in Texas collectively consume about 25 billion gallons of water each year. That volume is expected to soar to between 29 and 161 billion gallons by 2030, a wide range that reflects the significant gaps in publicly available data. 

“Texas’ State Water Plan does not include projected demand growth for data centers,” the HARC report states. “Because there are already unmet needs, including the current 4.8-million-acre-foot shortage as determined by the Texas Water Development Board (TWDB), data centers’ unknown unmet needs are poised to place unprecedented stress on local water supplies across the state.”

Each data center can “drink” as much as an entire community. Yet Texas does not require data center operators to disclose projected water use or report actual consumption. This makes research difficult and limits visibility for water-stressed municipalities facing rapid growth decisions.

“It is concerning that we have this explosive growth in data centers but no way to forecast that water use from a state or regional planning perspective,” Margaret Cook, Vice President of Water and Community Resilience at HARC, told the Texas Observer. “We have no way of reconciling what these cases will do to water supplies or existing plans because data centers don’t share their plans with the state or regional planning groups.”

Carlos Rubinstein, a former Texas water regulator who helped lead both the TWDB and the Texas Commission on Environmental Quality, says that policymakers rely on these regional planning groups to petition necessary amendments to the state water plan as new data emerges. “State water planning accepts the fact that changing conditions will take place at frequencies that are at times not aligned with the every 5-year plan review. The plan can easily be amended with well-reasoned, justified data that calls for updated demand projections and the identification of new water management strategies,” Rubinstein said in an email. 

However, without laws mandating such disclosure, no new data is emerging. Therefore, there is no way for planning agencies to project future demand or to provide long-term recommendations. Regional planning authorities now lack the grounds to even petition for amendments to the state water plan, experts say, creating a systemic vulnerability particularly in rural areas where a single data center can become the largest community water consumer overnight. 

“If a speculative developer is a bad actor, they could approach small communities who manage their own water supplies and see dollar signs from these data centers,” Cook said. “The communities could get a tax win for the area from the data center coming in, but without proper planning and information, they might be trading their water supply and putting their community at risk of future shortages, or at a much higher expense of future water supplies, in exchange for a short-term win.” 

Despite growing consensus about the scale of the risk, no state agency currently has clear authority to require data centers to disclose water use or prioritize non-potable supplies. Texas law treats data centers largely as commercial customers of municipal utilities rather than as heavy industry. This leaves oversight fragmented among local governments, water utilities, and regional planning groups.

That planning gap has begun to attract policymakers’ attention. Concerns about stressing aquifers already depleted by drought and population growth, as well as rising utility costs in economically disadvantaged communities, has led some to call for AI to emulate an unlikely industry: oil and gas. 

In December, the Texas GOP adopted a resolution demanding that the AI industry “follow the same water management and recycling protocols currently required of the oil and gas industry,” pushing the industry toward large-scale water recycling as freshwater supplies have diminished. Over the last two years, major oil and gas producers—most notably ExxonMobil—have reduced their reliance on freshwater in the Permian Basin, shifting the bulk of their operations to recycled “produced water,” which is non-potable water pushed to the surface during oil and natural gas extraction. 

“The company has made significant progress in reducing freshwater dependence in the region, increasing our use of recycled produced water in our hydraulic fracturing operations from 64 percent in 2022 to 87 percent in 2024,” ExxonMobil stated in an April 2025 sustainability report. The transition has significantly reduced oil and gas operators’ reliance on groundwater, thereby easing aquifer depletion.  

The abundance of produced water could effectively turn what was once a waste product into a large-scale industrial water supply. The GOP resolution also calls for AI data centers to “prioritize the use of recycled water from oil and gas operations to protect Texas aquifers.” 

Oil and gas operators were pushed toward reuse and data disclosure through state and federal regulation. Yet, no comparable requirements exist for data centers, even as their water demand scales to unprecedented levels. 

Data centers are being built faster than state water plans can be updated. Once contracts are signed and cooling systems requiring freshwater are built, the opportunity to require large-scale reuse disappears. By delaying AI water use and disclosure legislation, water experts warn, Texas risks locking in decades of freshwater demand before regulatory frameworks catch up. 

“The data center industry shares the obligation to pursue sustainable water development and use strategies,” said Rubinstein, drawing on his experience at the TWDB. “We have always known that the best water management strategy is based on water you already have.”

The post The Texas AI Boom is Outpacing Water Regulations appeared first on The Texas Observer.