人工智能算力高质量发展评估体系报告

Industry Investment Rating - The report does not explicitly provide an investment rating for the industry [1] Core Viewpoints - The report emphasizes the importance of high-quality computing power in driving the development of artificial intelligence (AI) and the digital economy [1] - It highlights the transition from quantity to quality in computing power development, focusing on efficiency, intelligence, and sustainability [1][13] - The report identifies key challenges such as insufficient computing power supply, low intelligence levels, and energy consumption issues [8][9][10] Development Status and Challenges Development Status - Policy: Governments worldwide are accelerating the construction of computing power competitiveness through policy support and strategic planning The US plans to invest over $2511 billion in AI-related fields, while China is promoting high-quality development of computing power infrastructure [5] - Technology: Generative AI technologies, such as ChatGPT, are rapidly advancing, driving the need for more powerful computing resources [7] - Market: Investments in computing power are increasing globally, with the US planning to invest $2800 billion in chip technology and China's "East Data West Computing" project attracting over 435 billion yuan in direct investment [8] - Scale: Global computing power is expanding, with intelligent computing power growing by 136% year-on-year, reaching 335 EFLOPS by the end of 2023 [12][13] - Development Level: China's computing power industry is shifting from scale expansion to quality improvement, focusing on application-driven development and green energy efficiency [13][14] Challenges - Challenge 1: Insufficient computing power supply and mismatched demand, with AI model training requiring massive computing resources [14] - Challenge 2: Low intelligence levels of computing power, making it difficult to meet diverse application scenarios [15] - Challenge 3: Energy consumption and carbon emissions are significant issues, with computing centers consuming 1500 billion kWh of electricity in 2023 [16] - Challenge 4: Rising demand for diverse computing power, but the level of universal accessibility and affordability remains low [17] - Challenge 5: Incomplete supply chains and underdeveloped ecosystems, with compatibility issues between different hardware and software platforms [18] - Challenge 6: Simple performance evaluation systems, lacking comprehensive assessment of actual computing power performance [19] Definition, Connotation, and Characteristics Definition - High-quality computing power is defined as advanced computing capability based on the latest AI theories, combined with algorithms and data, driving productivity and economic development [20][21] Connotation - Technological Innovation: High-quality computing power serves as the main engine for AI model training and application, reducing the threshold for AI adoption [23] - Optimization of Production Factors: It optimizes the allocation of data and resources, enhancing the efficiency of production, distribution, and consumption [24] - Industrial Transformation: It drives the integration of advanced technologies with traditional industries, fostering new business models and services [24] Characteristics - High Computational Efficiency: Focuses on both theoretical and practical performance, with an average computational efficiency of 118 GFLOPS/W in China [25][26] - High Intelligence Efficiency: Combines efficient AI processing with intelligent optimization capabilities [27] - High Carbon Efficiency: Aims to maximize computing output with minimal carbon emissions, emphasizing lifecycle carbon management [28] - Accessibility: Ensures computing power is widely available and affordable, supporting diverse application scenarios [29] - Sustainability: Emphasizes technical compatibility, complete supply chains, and open industrial ecosystems [30] - Evaluability: Requires a comprehensive evaluation system to reflect the actual performance of computing power [31] Development Path and Outlook Development Path - System Design: Focuses on improving computational efficiency through optimized system architecture and resource management [32] - Collaborative Drive: Enhances intelligence efficiency through the integration of computing power, algorithms, and data [33] - Lifecycle Management: Promotes carbon efficiency through green procurement, design, and operation [34] - Infrastructure First: Advances universal accessibility and affordability by building intelligent computing centers [36] - Ecosystem Prosperity: Encourages sustainable development through open and standardized industrial ecosystems [38][39] - Diverse Evaluation: Accelerates the standardization of computing power development through comprehensive evaluation systems [42][44] Outlook - Market Environment: China's vast application market and supportive policies provide a strong foundation for the computing power industry [46] - Universal Accessibility: The trend towards universal accessibility will unlock the potential of AI across various fields [47] - Intelligent Upgrade: The integration of AI and computing power will drive deep economic and social development [48] Evaluation System Exploration Background - The evaluation system for computing power is transitioning from hardware-focused assessments to comprehensive evaluations that consider application performance and lifecycle carbon management [51][52] Principles - The evaluation system should be policy-aligned, systematic, targeted, comprehensive, operable, and adaptable to future changes [58] Practice - The "Five-in-One" evaluation system assesses computing power quality across five dimensions: computational efficiency, intelligence efficiency, carbon efficiency, accessibility, and sustainability [60][70] Significance - The evaluation system standardizes and accelerates the high-quality development of the computing power industry, providing guidance for technological innovation and infrastructure construction [70] Application Recommendations - Accelerate the development of supporting standards and tools to ensure the effective implementation of the evaluation system [71] - Conduct evaluation tests in typical AI application scenarios and expand theoretical research to support the industry's high-quality development [73]