OpenAI

GOOG

GOOGL

<p>In the reality where public grid construction takes at least ten years, leading American AI laboratories such as OpenAI, xAI, and Google collectively bypass the grid and build their own gas power plants to get computing power running as quickly as possible. The underlying logic is that when AI enters the stage of ultra-large-scale deployment, the power issue has upgraded from a "cost issue" to the primary constraint determining whether computing power can go live on schedule</p>

- The increasing demand for computing power in AI is overwhelming the outdated U.S. public power grid, driving data centers to build on-site power plants.  
- A report by SemiAnalysis highlights that AI’s expansion has turned power supply timing into a critical constraint, as centers face mismatched timelines between quick data center constructions and lengthy grid upgrades.  
- Companies are adopting self-generated power strategies to mitigate the risk of delays, focusing on immediate deployment with solutions like gas turbines and engines, which could redefine how power is managed in the AI era.

SemiAnalysis in-depth report: The U.S. power grid can't keep up, AI data centers are racing against time to "build their own power plants"