谁等电网,谁就出局? 美国AI巨头掀起“自发电”热潮
Xin Lang Cai Jing·2026-01-02 03:04

Core Insights - A silent energy revolution is accelerating in the core of the U.S. AI industry, with leading AI labs like OpenAI, xAI, and Google opting to bypass the public grid by building their own gas power plants [1][3] Group 1: Energy Demand and Supply Mismatch - The report from SemiAnalysis highlights that electricity has transformed from a mere operational cost to a primary constraint determining whether computing power can be deployed on schedule [3] - AI data centers are facing a critical mismatch between the pace of grid delivery and the speed of computing power expansion, with construction timelines for data centers being 12-24 months compared to 3-5 years for grid expansion [3][4] - The Texas Electric Reliability Council (ERCOT) has seen data center load requests reach several tens of gigawatts, yet only about 1 gigawatt has been successfully integrated into the grid during the same period [3] Group 2: The Shift to On-Site Power Generation - AI companies are willing to incur higher costs for on-site power generation due to the "time value" of computing power, with a 1-gigawatt AI data center potentially generating annual revenues in the range of billions of dollars [4] - The "BYOG" (Bring Your Own Generation) model allows AI companies to quickly deploy facilities off-grid initially and later connect to the grid, with xAI's rapid construction of a supercluster in Memphis serving as a benchmark [4][5] - By deploying over 500 megawatts of fast-moving gas turbines and engines, xAI has demonstrated a commitment to speed, which is being widely emulated across the industry [4] Group 3: Industry Trends and Future Implications - By the end of 2025, on-site power generation is expected to become a systemic trend, with collaborations like OpenAI and Oracle building a 2.3-gigawatt gas power plant in Texas [5] - Natural gas has emerged as the dominant choice for on-site power generation due to its deployment speed, stability, and technological maturity, while alternatives like nuclear power and renewable energy sources face longer construction timelines [5] - The shift towards self-generated power signifies a new era where the ability to deliver electricity quickly may become as crucial as the quality of algorithms and chips in the competitive AI landscape [6]