大洋彼岸的“AI擦边”生意：废弃电厂变身提款机，矿机商拿到AI帝国入场券

Core Insights - The United States is facing an electricity shortage crisis driven by the increasing power demands of generative AI, with data center electricity consumption expected to double by 2030, reaching approximately 945 terawatt-hours, equivalent to Japan's annual electricity usage [1] - The aging infrastructure of the U.S. power grid, built over 40 years ago, is struggling to meet the sudden surge in energy demand from AI, leading to a mismatch between old systems and new energy needs [1] - The shift towards AI is reviving old peaker power plants, with about 60% of those scheduled for retirement by 2025 delaying their shutdowns due to increased electricity demand from data centers [3][5] Group 1: AI and Power Demand - The energy consumption of a typical ChatGPT query is approximately 2.9 watt-hours, nearly ten times that of a traditional Google search, highlighting the significant energy requirements of AI technologies [1] - The rapid growth of AI workloads has led to electricity demand from data centers exceeding the existing supply capabilities of the PJM grid, resulting in soaring electricity prices and power shortage alerts [3] - The reliance on renewable energy sources like wind and solar is insufficient for the continuous power needs of data centers, creating a "grid bottleneck" and a lack of baseload power [1] Group 2: Revitalization of Old Power Plants - Many peaker plants, designed to operate during peak demand, are being kept online despite their environmental impact, as they lack adequate pollution controls and are often located near low-income communities [7] - The costs associated with keeping these peaker plants operational have surged, with PJM paying over 800% more this summer compared to the previous year to ensure their availability [7] Group 3: Bitcoin Miners and AI Infrastructure - Bitcoin miners, previously seen as outsiders, are becoming key players in the AI infrastructure race due to their established power assets, particularly in Texas, where they have taken advantage of cheap land and electricity [8][9] - The transition of Bitcoin miners to AI data centers is driven by the deteriorating economics of mining, with many miners opting to reduce operations and shift towards more stable revenue from AI [9][13] - Texas has enacted legislation requiring new large data centers to switch to backup power during grid stress, reflecting the growing competition for limited electricity resources [13] Group 4: Investment Trends in Data Centers - Data center construction spending is projected to surpass that of office buildings as AI drives demand, with a potential $1 trillion investment in new data centers in North America from 2025 to 2030 [14][17] - Major real estate firms are increasing investments in data centers, with 95% of key investors planning to allocate more resources to this sector, moving away from traditional real estate [19] - The concentration of data center tenants among a few large cloud service providers poses significant risks, as delays in construction or power supply issues could trigger contract terminations [21][23]