Why Texas Is Making Data Centers Wait | Andreessen Horowitz
Short summary
Texas' grid operator hit pause on new data center permits after 474 GW of projects queued up—mostly speculative submissions that overwhelmed the interconnection system. The freeze reveals real problems: developers spam applications across multiple sites without firm commitments, communities get kept in the dark about resource impacts, and the grid's reliability rules make it expensive and slow to add massive new loads.
- Texas' grid queue exploded from 63 GW to 474 GW in six months (90% data centers), forcing ERCOT to pause approvals because planners can't distinguish real projects from speculative ones, and ratepayers would foot the bill for upgrades that never get used.
- Connecting a data center to the grid now requires expensive infrastructure upgrades since the system is stretched thin and can't handle simultaneous load drops—new rules force data centers to stay connected through power faults and charge them based on peak capacity, not actual usage.
- Developers are building on-site power plants to skip the grid queue entirely, but nearly all plan to connect to ERCOT eventually anyway since grid power costs less; the real bottleneck is the 5-10 year interconnection timeline, not the technology.
Full summary
Texas' power grid is getting swamped. By mid-2024, the state's grid operator had 474 GW of new customers waiting to connect—mostly data centers—which is five times the record peak demand the grid has ever handled. The state hit pause on new permits in response. The problem isn't just that there's too much demand; it's that the interconnection process is broken. Developers spam applications across multiple sites with speculative projects that may never materialize, forcing grid planners at ERCOT to waste resources studying phantom loads. Meanwhile, communities hosting these data centers get stonewalled. Of 377 companies surveyed about their resource use, only 28 answered. Officials in Hood County were asked to approve tax waivers for "Project Patriot" without knowing who was behind it—code names hide real identities during site shopping, but that excuse falls apart once tax breaks get voted on.
The technical challenge is that the grid has no spare capacity left. A decade ago, Amazon and other big power buyers could sign long-term contracts with utilities that had room to spare. Now almost any new large load requires expensive infrastructure upgrades—new wires, substations, sometimes entire power plants. Data centers are trying workarounds like colocation, where they build campuses next to existing power plants to draw power before it hits the grid. But regulators keep blocking this. When Amazon tried to buy power directly from Talen's nuclear plant in Pennsylvania, federal regulators rejected it because it would let Amazon dodge grid fees and shift transmission costs onto everyone else. Similar fights happened in Texas with CyrusOne and Crusoe Energy, where facilities got approved only if they agreed to cut power within 30 minutes if the grid calls for it—meaning they need backup systems to stay viable.
The real friction comes down to who pays for resilience. ERCOT studies what happens when transformers fail, lines go down, or massive loads suddenly switch to backup power (which happened in Virginia this year and nearly destabilized the grid). The grid is designed for worst-case scenarios—scorching summers and brutal winters where failures mean people die. To handle this, you need infrastructure sized for peak demand that sits mostly idle the rest of the year. Texas now proposes charging large customers based on their peak capacity needs year-round, not just what they actually use. But here's the catch: developers can commit to $100,000 study fees plus $50,000 per MW in deposits, which sounds substantial until you realize they only forfeit 20% if a project fails. If a 1 GW data center campus never materializes, the developer loses maybe $10 million while everyone else on the grid covers the cost of infrastructure built for it. The whole process can take 5 to 10 years, and right now ERCOT has frozen even that, waiting for an audit to sort out which projects are actually real.