当 AI 算力飞向太空：美国科技资本正在重走“苏联”的路

Core Viewpoint - The article discusses the emerging trend of relocating data centers to space as a potential solution to the limitations faced by the AI industry, highlighting the historical parallels with the Soviet Union's resource misallocation in technology development [1][3][12]. Group 1: Space as a Solution - The current AI boom has led to a misconception that the bottleneck lies in chip production, while the real constraints are energy and land availability [5][11]. - The construction of large data centers in the U.S. faces a "triple constraint" of lengthy power approval processes, land and environmental restrictions, and rising cooling system costs [5][11]. - Some companies are seriously considering extreme solutions, such as relocating data centers to space, to overcome these terrestrial limitations [5][6]. Group 2: Economic Viability of Space Data Centers - The feasibility of space data centers is questioned, with estimates suggesting that building a 1GW space data center could exceed $100 billion, significantly higher than ground-based alternatives [9][10]. - Current launch costs range from $1,500 to $3,600 per kilogram, and to be economically viable, these costs would need to drop below $300 per kilogram, requiring an 80% reduction [9][10]. - The extreme conditions in space necessitate advanced hardware that can withstand high-energy cosmic radiation, complicating maintenance and increasing costs [10][11]. Group 3: Historical Parallels and Industry Implications - The article draws parallels between the current U.S. tech capital trends and the Soviet Union's past, where resources were heavily invested in grand projects at the expense of commercial efficiency [12][14]. - The focus on massive infrastructure projects in AI, such as space data centers, signals a potential decline in marginal returns and a shift towards "national engineering" rather than commercial viability [14][15]. - The discussion of relocating data centers to space may indicate that the growth potential in the terrestrial realm is becoming insufficient, marking a critical point in the AI narrative [15].