The AI Bottleneck Isn't Energy Supply—It's the Electric Grid

The massive investment in AI computing projects like Stargate requires enormous electricity, but faces significant grid interconnection bottlenecks. While energy supply is not the primary constraint, connecting new data centers to the electric grid is severely backlogged, limiting future AI model training capabilities.

The Scale of the Challenge

The Stargate project in Abilene, Texas, will cost over $40 billion and draw 1.2 gigawatts at peak load. Total AI computing power could reach 100 gigawatts worldwide by 2030 if 2025 growth rates continue.

Grid Interconnection Backlogs

The grid interconnection process has become significantly backlogged, increasing wait times from less than 20 months in 2005 to 55 months by 2023. Major AI company leaders agree that future data centers will be power-limited due to grid constraints.

Market Signals and Energy Sources

Electricity prices and market mechanisms signal developers to balance different energy sources based on supply and demand. Solar overproduction in ERCOT reduced the value of midday power from 92.9% to 38.7% between 2020 and 2025.

Infrastructure Development

ERCOT battery storage capacity grew from almost none in 2020 to 8.6 gigawatts by October 2025. The abundance of [AI] will be limited by the abundance of energy, according to industry leaders who say they would build bigger clusters if they could get the energy to do it.

We need capacity – a lot of capacity.