Why Texas Is Making Data Centers Wait

Texas has become the epicenter of a collision between AI's power demands and grid reality. At the end of 2024, the state's grid operator had 474 GW of new customers waiting to connect—mostly data centers—compared to peak demand of about 90 GW. That explosion forced regulators to hit pause on new permits and demand that data centers fund their own grid upgrades. The underlying problem isn't just the sheer volume of requests. It's that developers flood the system with speculative projects across multiple sites, many from companies that have never actually built a power plant. ERCOT, which runs most of Texas's grid, can't distinguish real projects from noise, and ratepayers end up footing the bill for infrastructure built for developments that never materialize. Local communities are also pushing back—they want straight answers about noise, water use, and emissions before tax breaks get handed out, but only 28 of 377 companies bothered answering a state survey about their resource use. Connecting a data center to the grid is far more complicated than it was a decade ago. When Amazon could just sign long-term power contracts without proving the electricity could physically reach them, things were simple. Now the grid is maxed out, so almost any new large load requires expensive upgrades to wires and substations. Developers have tried workarounds like building next to existing power plants to draw power before it hits the grid—a setup called "colocation"—but regulators keep tightening the rules. Amazon's campus beside a nuclear plant in Pennsylvania got rejected for direct power delivery. A CyrusOne campus in Texas had to agree to cut its power use within 30 minutes if the grid needs it. Two Crusoe data centers sharing a wind farm had to accept rules requiring them to shed loads that exceed what the wind farm can actually produce. The core issue: interconnection studies that used to handle 50 large loads now face 225 requests, forcing ERCOT to batch-process approvals and figure out which infrastructure upgrades are actually needed. The real bottleneck is cost allocation and risk. Texas charges large customers based on their electricity demand during summer peak hours, so a data center that ramps down on the hottest days avoids most of its transmission bill. Regulators proposed charging based on 12 monthly peaks instead, forcing large loads to pay proportionally to their pipe size rather than actual usage. The deposit to reserve grid capacity is surprisingly low—$100,000 flat plus $50,000 per megawatt—and even if a project fails, the developer forfeits only 20% of the deposit while everyone else absorbs costs for regional upgrades. A full grid connection takes 5 to 10 years, so phased approaches are becoming standard to give utilities gradual targets and let developers prove they can deliver. Right now, even that isn't moving: ERCOT has paused approvals for data centers 75 MW or larger while it figures out how to handle Batch Zero's workload.