Artificial intelligence is often governed through rules directed at models, data, safety, or automated decisions. Texas is demonstrating another form of AI governance: regulation of the physical infrastructure required to make large-scale computation possible.
On September 21, Governor Greg Abbott directed the Texas Commission on Environmental Quality to halt permits sought by data centers until the Electric Reliability Council of Texas completes an audit process. The directive followed earlier action requiring the Public Utility Commission of Texas and ERCOT to examine large data-center projects seeking grid interconnection. State officials have also increased pressure on large water users to comply with reporting requirements. The immediate dispute concerns permits, electricity, water, and infrastructure costs. The broader issue is public authority over the material systems on which AI expansion depends.
Compute becomes a public-resource question
Data centers convert investment in artificial intelligence into demands on land, electricity networks, water systems, transmission capacity, backup generation, and local infrastructure. These demands are geographically concentrated. A model may appear as a digital service to its users while its computational infrastructure produces physical consequences for communities and utilities located far from those users.
Texas is especially important because its scale, energy market, available land, and rapid development have made the state attractive to data-center operators. The same conditions expose institutional limits. An individual interconnection proposal can be evaluated as a discrete project. Hundreds of projects create aggregate effects that cannot be understood by reviewing each facility in isolation. Grid planners must estimate coincident demand, transmission requirements, reliability obligations, and the probability that proposed projects actually materialize.
Permitting becomes coordination
The September directive links environmental permitting to the completion of energy and resource audits. That linkage is analytically significant. It treats data-center development as a cross-agency coordination problem rather than a collection of separate permits. Electricity regulators, grid operators, water agencies, and environmental authorities need information from one another before any single institution can determine whether a project is sustainable.
This form of governance changes the sequence of private investment. Developers can no longer assume that access to capital and a suitable site are sufficient. Public institutions can make interconnection, water reporting, environmental authorization, and infrastructure-cost allocation conditions of development. In practical terms, the state is deciding which costs remain private and which may be shifted to ratepayers, communities, or future infrastructure budgets.
Who pays for AI infrastructure?
Abbott's formulation that data centers must "pay their own way" captures one of the central political-economy questions surrounding AI infrastructure. New industrial loads can justify transmission investment and generate tax revenue, employment, and local economic activity. They can also create network costs that are socialized through utility systems. The distribution of those costs depends on rate design, interconnection rules, infrastructure-financing arrangements, and negotiated incentives.
The same issue applies to water. Cooling technologies vary substantially, and water demand differs across facilities and operating conditions. Governance therefore requires evidence rather than generalized claims about whether data centers are inherently water intensive. Reporting rules matter because public institutions cannot allocate scarce resources or assess cumulative demand without reliable facility-level information.
AI governance moves into ordinary institutions
The Texas case shows why AI policy cannot be confined to specialist technology regulators. Public utility commissions, environmental agencies, water boards, grid operators, zoning authorities, tax offices, and local governments increasingly shape the pace and geography of AI development. Their decisions determine where computing capacity can be built and what obligations accompany access to public infrastructure.
This institutional migration has consequences for democratic accountability. Communities may encounter AI expansion through electricity bills, water restrictions, tax incentives, transmission projects, or local land-use decisions rather than through debates over model regulation. The political legitimacy of AI infrastructure will therefore depend partly on whether those institutions can demonstrate that benefits and burdens are allocated through transparent rules.
From industrial boom to governed expansion
Texas has not rejected data-center development. Its current policy represents an attempt to make growth conditional on verification and system-level planning. Whether the approach succeeds will depend on how audits define capacity, how regulators distinguish speculative projects from credible demand, how costs are assigned, and whether permitting standards remain stable enough for long-term investment.
The larger lesson is that compute is becoming governed infrastructure. As AI systems scale, their physical requirements bring them into older domains of public authority. Electricity, water, environmental permits, and transmission are becoming instruments through which states shape the development of artificial intelligence. The future geography of AI may therefore be decided as much in utility dockets and resource audits as in laboratories and technology companies.