OpenAI told Texas Gov. Greg Abbott it will cover the infrastructure costs generated by its projects, support new power generation and limit water use, responding days after Abbott froze data center approvals pending a statewide audit.
The August 7th letter, published Monday, was signed by Uday Ruddarraju (@udayruddarraju), OpenAI's newly appointed CTO of compute. Ruddarraju joined OpenAI in 2025 after infrastructure roles at xAI, Robinhood and eBay. In announcing his new position, he described OpenAI's goal as building the world's largest compute footprint, with work spanning distributed systems, hardware, manufacturing and data center construction.
That expansion depends on state officials allowing OpenAI and its infrastructure partners to keep adding power-hungry campuses. Texas has become central to OpenAI's compute strategy, and Abbott's August 3rd directive placed every project moving through the state's grid interconnection process behind a new regulatory checkpoint.
OpenAI adopts Abbott's infrastructure test
OpenAI's letter tracks the categories in Abbott's audit nearly point for point. OpenAI said its projects will fully fund project-driven infrastructure costs rather than shift them to residential or small-business customers. OpenAI also committed to working with utilities, the Electric Reliability Council of Texas and infrastructure partners to support the generation and grid equipment needed for its growth.
On water, OpenAI said it will minimize consumption and prioritize efficient cooling, including closed-loop systems. OpenAI also promised to address noise, light, traffic, setbacks, land use and emergency response through protections designed for each site.
OpenAI further offered to disclose information about electricity and water use, infrastructure spending, public incentives and community protections to state and local officials.
The commitments remain broad. The letter sets no project-level targets for water consumption, new generation or infrastructure spending. Its power pledge says OpenAI will "support" the resources needed for its growth, while its cost commitment applies specifically to "OpenAI-developed projects." OpenAI also promises transparency around public incentives without pledging to forgo them.
Those distinctions will matter as regulators assess projects involving separate developers, utilities, cloud providers and capital partners. OpenAI's Texas footprint is built through such arrangements rather than through OpenAI alone.
Texas is already running OpenAI workloads
OpenAI said in January that its first Stargate site in Abilene was already training and serving frontier AI systems. OpenAI has committed to securing 10 gigawatts of US AI infrastructure by 2029 and said at the start of 2026 that planned capacity had already passed the halfway point.
OpenAI and SoftBank also each invested $500 million in SB Energy in January. As part of that agreement, OpenAI signed a 1.2-gigawatt lease for a Milam County data center. OpenAI and SB Energy said that campus was designed to minimize water use and would be paired with new generation intended to protect Texas ratepayers.
Those earlier statements align with the commitments in Ruddarraju's letter. Texas regulators will now determine whether individual projects can document them with specific ownership, consumption, financing and mitigation plans.
Abbott's audit stops the queue
Abbott's August 3rd order requires the Public Utility Commission of Texas and ERCOT to audit all data centers advancing through the grid interconnection process before any project moves forward. Applicants must disclose expected tax assistance, peak and annual electricity demand, power-generation plans, water requirements, cooling methods, community protections and controlling interests.
ERCOT was considering more than 474 gigawatts of connection requests when Abbott issued the directive, according to the governor's office. Abbott said about 90% of the requested load came from data centers. The queue represents prospective demand rather than capacity certain to be built, but its scale is more than five times ERCOT's record peak demand.
The order followed weak participation in a state survey intended to measure data center power and water use. Of 377 companies contacted by Public Utility Commission staff, 28 submitted responses, The Texas Tribune reported. Abbott had already directed regulators on June 10th to prevent data center infrastructure costs from reaching residential customers and called for legislation addressing water-efficient cooling, usage reporting, tax incentives, setbacks and noise.
OpenAI's response positions Ruddarraju's compute organization as an early volunteer for Abbott's framework. The practical test will come in the filings: how much new generation OpenAI-backed campuses bring, who pays for transmission and interconnection work, how much water each site consumes, and which incentives OpenAI and its partners receive.