Works in Progress argues that AI data-center growth is being slowed by grid interconnection, not a simple lack of electricity, using Stargate in Abilene, Texas and other high-load projects as examples. The article says Stargate is expected to draw 1.2 gigawatts at peak load and cites an EPRI/Epoch AI projection that AI computing power could reach 100 gigawatts worldwide by 2030 if 2025 growth continues. For infrastructure teams, the key planning variable is now queue time: utility studies, transmission upgrades, and local permitting can delay a campus even when financing and GPUs are available. That makes power strategy, backup generation, and site selection first-order model-deployment constraints.
The practical takeaway for AI infrastructure teams is that compute planning now depends on utility process design. GPUs, capital, and land do not translate into deployed capacity if interconnection studies, transmission upgrades, and local grid constraints push energization years into the future.
What happened
Works in Progress argues that the AI buildout is bottlenecked by getting electricity to data centers, not by a simple shortage of electricity. The article uses Stargate in Abilene, Texas as a flagship example, saying the campus is expected to draw 1.2 gigawatts at peak load. It also cites an EPRI/Epoch AI projection that global AI computing power could reach 100 gigawatts by 2030 if the 2025 growth rate continues.
Technical context
The core constraint is interconnection. Grid operators need to model how new loads affect power flows and whether transmission or substation upgrades are required. Works in Progress says median power-plant interconnection waits rose from less than 20 months in 2005 to 55 months by 2023, illustrating why large AI campuses can be delayed even when generation exists somewhere on the grid.
For practitioners
Capacity planning should treat power delivery as a schedule risk, not an external facilities detail. Model-training and inference roadmaps need assumptions for queue position, backup generation, energy contracts, cooling, and whether nearby fabs or battery plants are competing for the same transmission capacity.
What to watch
The next useful signals are utility interconnection reforms, data-center self-generation deals, and whether Stargate-scale campuses can secure enough reliable power without creating regional bottlenecks for other industrial loads.
Key Points #
- 1Grid interconnection queues can delay AI campuses even when capital, land, and GPU supply are available.
- 2Large fabs, battery plants, and data centers may compete for the same transmission upgrades and utility study capacity.
- 3Infrastructure teams should model power delivery and energization dates as first-order constraints in AI deployment plans.
Scoring Rationale #
This is notable because grid interconnection has become a first-order bottleneck for AI infrastructure timelines and cost planning. The score is slightly below major because the event is analytical rather than a new regulatory action, financing event, or confirmed project delay.
Sources #
Public references used for this report. Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.