Greg Abbott freezes Texas grid access for AI data centers
- Texas froze grid connections after 1,800 AI proposals requested 474 gigawatts of electricity.
- Wall Street launched a $500B private credit framework to finance off-grid compute hardware.
- AI developers are deploying off-grid solar and battery plants to bypass four-year utility queues.
The physical wall hit Texas first. Governor Greg Abbott ordered an immediate halt on approving new grid connections for data centers after requests reached 474 gigawatts - more than five times the state’s peak record.
On This Week in AI, the numbers revealed why regulators panicked. Data centers account for 90 percent of all new power requests in Texas, across 1,800 separate proposals. Following memory of disastrous winter blackouts, state officials refused to let speculative compute facilities overwhelm municipal power supply. Tax GPT CEO Kash Ali observed that if low-regulation, land-rich Texas cannot absorb the buildout, developers face a wall with few domestic alternatives.
The next day on All-In, the conversation shifted from regional grid bottlenecks to systemic economic friction. Gavin Baker warned that expanding AI revenue requires massive energy production that old-world supply chains cannot deliver. Turbine blade manufacturers currently operate 24-hour shifts using 40-ton presses, yet backlogs continue to grow, prompting developers to strip turbine engines off retired private jets for off-grid power.
To bridge the infrastructure gap, Wall Street is turning hardware into structured debt. Nvidia partnered with BlackRock, KKR, and Goldman Sachs to establish a $500 billion debt financing framework. Private credit funds now lend directly to data center operators, securing loans against Nvidia graphics processors and future token revenue. As David Sacks noted, institutional lenders now treat processing capacity like income-generating real estate.
Two days later on Moonshots, investor Ramez Naam laid out the core disconnect facing the industry. Chip makers will supply over 200 gigawatts of AI compute capacity by 2030, but U.S. utility buildouts will barely deliver 100 gigawatts. Connection wait times for transmission lines have surged from 15 months to nearly four years.
Monopoly utilities lack financial incentives to accelerate construction under cost-plus regulatory models. To bypass years of administrative delay, hyperscalers are deploying behind-the-meter natural gas generation and battery storage. In Texas, new rules allow data centers to bypass connection queues only if they operate as interruptible loads that draw power during off-peak hours.
Continuous solar-battery plants are emerging as the cheapest path around grid gridlock. Naam pointed to a gigawatt-scale facility in Dubai combining five gigawatts of solar panels with 19 gigawatt-hours of battery storage to deliver steady baseload power. At six dollars per watt, the solar-battery installation costs less than half of recent U.S. nuclear builds and deploys in under 12 months.
Because software workloads are mobile, capital is beginning to flee grid-locked jurisdictions altogether. Countries with abundant land, high solar exposure, and modernized laws - such as Australia, Mexico, and Middle Eastern states - are positioned to host the next generation of compute. Whichever nation eliminates grid friction first will capture the physical backbone of artificial intelligence.