Texas data centre developers are prioritising power infrastructure availability as the primary site selection criterion for AI deployments, shifting away from traditional facility-first approaches.
Artificial intelligence is forcing the data center industry to confront a fundamental constraint: compute can be deployed faster than the infrastructure required to power it. As AI and high-performance computing campuses move from hundreds of megawatts toward gigawatt-scale development, time to power has become one of the defining considerations in data center site selection, large-load interconnection and digital infrastructure investment.
In June, ERCOT reported tracking more than 438 GW of large-load requests, with nearly 89% associated with data centers. The scale of prospective demand has prompted ERCOT and the Public Utility Commission of Texas to implement new processes for evaluating large loads and their potential effects on the transmission system.
The shift is creating an important distinction between transmission infrastructure and deliverable power. A high-voltage transmission line near a development site can be an important infrastructure advantage, but its presence alone does not establish that hundreds of megawatts can be delivered on the schedule required by an AI or hyperscale customer.
For developers evaluating campuses requiring 250 MW, 500 MW, 1 GW or more, site selection increasingly involves broader questions: how much power can actually be delivered and when, what transmission upgrades may be required, whether generation can be developed alongside the load, and how fuel, water and fiber infrastructure will support phased development.
"The definition of a data center site is changing," said Roxanne Marquis, founder of 8888CRE.com. "For very large AI and high-performance computing projects, land and power can no longer be evaluated independently. Developers increasingly have to understand the entire infrastructure system required to turn a site into operating compute capacity."
This shift is contributing to increased interest in power-first development strategies. Rather than beginning with land and addressing power later, a power-first approach evaluates land, utility interconnection, transmission, generation, fuel, water and fiber as interconnected components of the development strategy.
Utility-supplied electricity remains central to long-term data center development. However, onsite and behind-the-meter generation are increasingly being evaluated as potential components of phased power strategies while long-term utility infrastructure is developed. ERCOT's current large-load framework recognizes configurations involving large loads that bring generation, including Withdrawal-Limited Private Use Networks. At the same time, Texas is increasing scrutiny of the broader effects of large data centers, including grid dependency, onsite generation, water consumption, cooling technology and community impacts.
These developments could broaden the geographic boundaries traditionally used for data center site selection. Established data center markets continue to offer significant advantages, including dense fiber networks, experienced contractors, existing customers and mature infrastructure ecosystems. But as individual projects reach hundreds of megawatts or more, access to scalable energy infrastructure can materially change the relative importance of geography. A site outside an established data center cluster may warrant consideration when contiguous acreage can be combined with transmission, grid capacity, generation potential, fuel infrastructure, water, fiber connectivity and an executable pathway to power.
For large AI data centers or hyperscale data center campuses, site acquisition is increasingly only the beginning of the development process. A property may be shovel-ready from a traditional real estate perspective and still require substantial electrical infrastructure development before it can support high-density compute or GPU infrastructure. Data center development and site development are becoming increasingly tied to power availability, time to power and time to market. The most consequential sites may be those where land, transmission capacity, power delivery, fiber diversity, cooling infrastructure and the broader development strategy can advance together.
"The question is no longer simply whether power infrastructure is near the property," Marquis said. "The more important question is whether the site provides a credible pathway to deliver the required power when the customer needs it."
At gigawatt data center scale, this analysis moves deeply into power systems engineering. Large-load interconnection can require transmission planning, load-flow analysis, power-flow studies, system-impact studies, short-circuit analysis and contingency analysis to understand whether the surrounding electrical system can reliably serve the proposed load. Questions involving N-1 reliability, substation and switchyard configuration, high-voltage transmission capacity and required network upgrades can determine whether apparent grid capacity ultimately becomes deliverable capacity.
In ERCOT, developers must also navigate the applicable interconnection process, transmission service provider requirements and large-load study procedures, including a Large Load Interconnection Study (LLIS) where applicable. The transmission service provider (TSP) and ERCOT each have important roles in determining how a proposed large load interacts with the transmission system. Understanding those requirements early can materially affect site acquisition, infrastructure planning, development risk and project schedules.
The growing focus on speed to power is also expanding the role of power generation development in data center site selection. Depending on the project and regulatory structure, developers may evaluate bridge power, onsite power generation, behind-the-meter power, dispatchable generation, natural gas generation or other generation technologies alongside the long-term utility solution. A Private Use Network (PUN) may also become relevant to certain configurations. Each approach introduces its own engineering and commercial requirements, including gas pipeline capacity, generation equipment, electrical interconnection, grid reliability, emissions and permitting, fuel contracting and infrastructure execution.
The objective extends beyond simply installing gas-fired generation to determine whether generation, transmission and utility power can form a credible phased power strategy that reduces development risk while accelerating the path to usable compute capacity.
The same standard increasingly applies to natural gas, water and fiber. Regional infrastructure should not be confused with infrastructure that has been engineered, contracted and demonstrated to be deliverable to a specific development. Gas pipeline capacity must be evaluated against the requirements of proposed generation. Fiber connectivity must ultimately address capacity and fiber diversity. Water requirements depend substantially on the cooling infrastructure and technology selected for the campus. Each component can affect infrastructure execution, capital requirements and the schedule between site control and operating compute.
As power moves further upstream in data center site selection, the boundary between engineering and development is beginning to narrow. Some of the most consequential work can occur before a site becomes a project, when engineers, developers and real estate teams are still determining whether the physical and electrical infrastructure can support the intended load. This creates an unusually interesting position for power systems engineers, interconnection engineers and transmission planning engineers who are drawn to difficult, undefined problems and want to work closer to where opportunities originate. At this stage, technical insight can do more than engineer a solution; it can help identify opportunities, expose development risk and determine whether a viable project exists.
For hyperscalers, AI infrastructure companies and data center developers, the result may be a broader approach to site selection. AI compute and high-density GPU infrastructure do not eliminate traditional requirements for location, connectivity, workforce and development feasibility. They can, however, make the ability to assemble sca