智算中心“狂飙”背后部分算力闲置

Core Insights - The rapid development of computing power in China has led to increasing concerns about idle capacity since the second half of last year [1] - According to IDC, the utilization rate of enterprise-level general computing power centers is only between 10% and 15%, with some intelligent computing centers showing GPU utilization rates below 30% [1] - Industry experts believe that while the situation of idle computing power is expected to improve this year, there is still a need to explore more efficient and reasonable new models for computing center construction [1] Group 1 - One reason for the idle computing power is the deviation in demand forecasting and the immature ecosystem [1] - The Chief AI Officer of Softcom Power, Jin Yadong, stated that despite the explosive growth in intelligent computing, the demand for AI applications is still in the early stages of development [1] - Although the number of AI applications has increased in recent years, it has not been sufficient to fully absorb the constructed computing power [1] Group 2 - The emergence of phenomenal AI products like DeepSeek has shifted market demand for computing power from training to inference [1] - There is an exponential growth in demand for inference computing power, leading to a mismatch between supply and demand as some established intelligent computing centers struggle to adapt their hardware and software structures to new requirements [1]