The True Bottleneck of the AI Expansion

By Works in Progress & Chris Gillett | June 25, 2026

The United States does not necessarily lack the electricity required to fuel its data center ambitions; rather, the crisis lies in the logistics of delivery—getting that power to the specific locations where it is needed.

The Scale of the Ambition

In Abilene, Texas, one of the most capital-intensive undertakings in human history is currently underway. Known as Stargate, this joint venture between OpenAI and Softbank is a high-performance computing campus designed to train the next generation of AI.

Energy Requirements at a Glance

The financial and energetic scale of these projects is staggering. Stargate alone is projected to cost over $40 billion.

The Interconnection Crisis

The primary obstacle to this growth is not the generation of power, but the interconnection process. Before any facility can plug into the grid, operators must conduct exhaustive studies to see how the new load affects power flows and determine if the existing infrastructure requires upgrades.

The Queue Problem

The timeline for these connections has deteriorated significantly: $\text{Median Wait (2005)} < 20 \text{ months} \implies \text{Median Wait (2023)} = 55 \text{ months}$

The current system is plagued by:

1. Dynamic Prioritization $\rightarrow$ Inflexible first-come, first-served queues, where critical projects are trapped behind low-value ones.
2. Rigid Evaluation Criteria that fail to incentivize plants that can self-manage their power needs during peak intervals.

A Future Defined by Electrons

The leaders of the AI industry are sounding the alarm. The consensus is clear: energy is the ultimate ceiling.

"Every data center in the future will be power-limited." — Jensen Huang, CEO of Nvidia

"Meta would build... bigger [AI training] clusters... if we could get the energy to do it." — Mark Zuckerberg, CEO of Meta

"The abundance of [AI] will be limited by the abundance of energy." — Sam Altman, CEO of OpenAI

The Macro Trend: Total Electrification

Beyond AI, the entire economy is shifting toward electricity because it is more efficient than combustion. Electricity converts to work instantly with minimal loss, which is why EVs are cheaper to fuel and maintain than internal combustion engines.

Between 1990 and 2024, we saw a precipitous drop in the cost of key components:

• Electric Motors: $\downarrow 97.5\%$
• Power Electronics: $\downarrow 99.5\%$
• Processors: $\downarrow 99.9\%$
• Batteries: $\downarrow 98.8\%$ (with a $5\times$ increase in energy density per kg)

This technological deflation has enabled a cascade of innovations:

• Portable electronics (Walkman $\rightarrow$ iPod $\rightarrow$ iPhone)
• Electric transit (Cars, vans, bicycles)
• Autonomous systems (Drones, robotic mowers/vacuums)
• Commercial humanoid robotics (Upcoming)

Grid Congestion and Regional Strain

This surge in demand is pushing existing grids to their breaking point. "Congestion" occurs when cheaper power is available but cannot reach the destination due to a lack of transmission cables, forcing the use of expensive local plants.

• Financial Impact: In 2023, US grid inefficiencies cost $11.5 billion (a 45% year-over-year increase).
• ERCOT (Texas): Predicts a power deficit by the summer of 2028.
• PJM (Mid-Atlantic/Midwest): Failed to secure enough future generating capacity in 2025.
• MISO (Midwest): Warns of growing "resource adequacy risks" without new capacity.

The Core Conclusion

While public discourse often focuses on the source of the energy—debating the merits of $\text{Nuclear vs. Solar vs. Gas}$—the more fundamental issue is transmission. The ability to move electricity efficiently is the true linchpin of the AI revolution.

{
  "status": "critical",
  "bottleneck": "transmission_and_interconnection",
  "impacted_sectors": ["AI", "Semiconductors", "EVs", "Robotics"],
  "primary_metric": "interconnection_wait_time"
}