中国AI算力行业发展报告

Industry Overview - GPU dominates the AI chip market with a 90% share in China's accelerator chip market in H1 2023, while non-GPU chips hold 10% [2] - The market size of China's accelerator chips is expected to grow from over 500,000 units in H1 2023 to over 900,000 units in H1 2024, with GPU share decreasing to 80% and non-GPU share increasing to 20% [2] - Domestic AI chip brands account for 10% of the market in H1 2023, expected to grow to 20% in H1 2024 [2] AI Chip Development - Google's TPU development timeline shows continuous advancements from TPU v1 in 2015 to TPU v5p in 2024, focusing on cost-effective inference and AI model training [4] - The evolution of TPUs has enabled large-scale training and inference capabilities, supporting AI model development [4] Data Center and Computing Power - China's data center capacity exceeds 8.3 million standard racks by June 2024, with computing power reaching 246 EFLOPS (FP32), growing at over 65% YoY [6] - Intelligent computing centers are rapidly expanding, with a focus on high power density (20-100kW per rack) and energy efficiency [5] - The AI server market in China grew from RMB 14.9 billion in 2020 to RMB 69.2 billion in 2023, projected to reach RMB 143.3 billion by 2028 [14] Cloud and AI Integration - China's cloud computing market reached RMB 616.5 billion in 2023, growing at 35.5% YoY, with expectations to reach RMB 2.14 trillion by 2027 [18] - Cloud providers are increasing AI computing power investments to meet the demands of large model training and inference [33] - AI cloud services offer advantages such as pay-as-you-go pricing, low-latency inference, and flexible resource deployment [21] Key Players and Innovations - Zhonghao Xinying, a domestic AI chip company, has developed the "Chana" TPU architecture, which outperforms international GPU chips by 1.5x in AI/ML scenarios [30] - The company's "Taize" AI computing cluster supports over 1,024 chips with high-speed interconnects, achieving performance tens of times higher than traditional GPU clusters [30] - Qingyun Technology has partnered with the National Supercomputing Center in Jinan to optimize computing resource management, achieving efficient scheduling of over 3,000P computing power [26] Market Trends and Applications - The global AI application market is booming, with downloads increasing by 26% YoY to 2.2 billion in the first eight months of 2024, and revenue growing by 51% to $2 billion [64] - AI applications are expanding across various sectors, including autonomous driving, smart cities, and industrial optimization [67] - The automotive industry is integrating AI models into vehicles, with companies like NIO, XPeng, and Li Auto developing their own AI platforms [86] Green Computing and Sustainability - China's data centers are focusing on reducing PUE (Power Usage Effectiveness) to meet national and local energy efficiency standards [90] - AI-driven energy optimization techniques are being implemented to improve resource allocation and cooling efficiency in data centers [91] - The integration of renewable energy sources and advanced cooling technologies is becoming a key focus for sustainable computing [100]