AI算力从“堆硬件”走向“拼效率” 产业链企业合力破解算网融合协同难题

Core Insights - The rapid expansion of AI computing power in China is highlighted, with the current capacity reaching 3000P for training and 1000P for inference, and over 250 innovative solutions developed across various sectors [1] - The integration of computing power and network technology, referred to as "算网融合" (computing-network integration), is identified as a critical path for addressing the growing demand for AI computing resources and enhancing computational efficiency [3][6] - The Chinese intelligent computing power market is projected to grow significantly, with estimates indicating a market size of $25.9 billion by 2025, reflecting a 36.2% increase from 2024 [2] Industry Growth and Trends - The intelligent computing power scale in China is expected to reach 1037.3 EFLOPS by 2025, representing a 43% increase from 2024, and is projected to double by 2026 [2] - The compound annual growth rate (CAGR) for China's intelligent computing scale from 2023 to 2028 is forecasted at 46.2%, indicating a strong trend towards large and super-large intelligent computing centers [2] - The demand for AI models is driving a significant increase in computing power requirements, with current needs exceeding hardware supply by over 200 times [3] Technological Developments - Companies are focusing on transforming data centers into intelligent computing centers, emphasizing the need for efficient, green, and high-performance solutions [4][5] - Huawei is leveraging advanced technologies such as zero-loss networking and intelligent computing network scheduling to enhance computing resource utilization from 40% to 75% [5] - The integration of distributed computing architecture, low-latency networks, and virtualization technologies is essential for achieving high-quality development in computing power [6] Market Dynamics - The competition among enterprises for intelligent computing resources is intensifying, with a focus on optimizing computing infrastructure and enhancing service capabilities [5] - The shift from hardware-centric approaches to efficiency-driven models in AI computing is becoming evident, necessitating a demand-oriented and benefit-oriented approach in intelligent computing center construction [6]