AI算力爆发式增长 数据中心面临哪些挑战？专家这样说

Core Insights - The rapid development of artificial intelligence (AI) technology has led to an explosive growth in computing power demand, with model parameters increasing from 117 million in GPT-1 to trillions in GPT-5, necessitating a shift in data center infrastructure to high-density, liquid-cooled, and clustered designs [1][4][5] Group 1: AI Infrastructure Evolution - The power density of single cabinets is increasing from 50 kW to 300 kW or higher, with supernode clusters requiring hundreds of MW, far exceeding the current data center capacity of 10-20 MW per building [4][5] - The transition from air cooling to liquid cooling is becoming essential as air cooling solutions approach their limits, necessitating simultaneous upgrades to data center infrastructure [5][6] Group 2: Design and Planning Recommendations - The "AIDC Room Reference Design White Paper" suggests creating an efficient power supply system, ensuring sufficient building height and load-bearing capacity for heavier AI servers, and optimizing interconnect designs for minimal latency [5][6] - The introduction of modular planning and compatibility between air and liquid cooling systems is recommended to facilitate the transition from low-density to high-density setups [6] Group 3: Domestic vs. International Development - International projects have achieved power densities of 120-150 kW per cabinet, with Nvidia planning to reach 600 kW by 2028, while domestic setups are primarily in the 40-60 kW range [7] - There is a notable difference in cooling strategies, with international projects utilizing lower temperature water for GPU efficiency, while domestic systems operate at higher temperatures for energy conservation [7] Group 4: Industry Standards and Collaboration - The Global Computing Coalition (GCC) has initiated the development of standards across various fields, with 20 standards already published and expectations to reach 30 by the end of the year [8] - The industry is committed to investing in new technologies to address challenges related to heat, electricity, and space in AIDC, while also enhancing collaboration and ecosystem development [8]