盘古智库发布《中国算力与能耗研究报告暨2030年发展预测》报告

Introduction - Computing power is recognized as a key infrastructure supporting the development of the digital economy and a core driver of new productive forces. The current scale and energy consumption of computing power in China show significant discrepancies among various research institutions and media [1] Research Methodology and Content - The report categorizes types of computing power and data centers, focusing on the statistics and calculations of intelligent computing power. It employs a unified FP16 precision for measurement while considering the relationship with FP8. The statistical methods include chip shipment calculations, server and cabinet conversions, and power reverse calculations to enhance measurement accuracy [1] Findings and Explanations - The report finds that if calculated based on the 8.3 million standard cabinets published by the government, the theoretical annual electricity consumption would be approximately 272.6 billion kWh, while the actual annual consumption is about 116.6 billion kWh, indicating a significant gap. This discrepancy challenges the narrative of excessive energy consumption in computing centers [1][2] Predictions for the 14th Five-Year Plan - The report predicts that during the 14th Five-Year Plan, driven by large models and generative AI, the scale of computing power in China will continue to expand, contributing to high-quality development in the AI industry [3] Computing Power Growth Forecast - It is expected that the annual growth rate of computing power in China will remain above 35% during the 14th Five-Year Plan, with the total computing power scale projected to exceed 2500 EFLOPS (FP16) by 2030. If measured using FP8, the total scale could surpass 5000 EFLOPS by 2030 [4] Energy Consumption Forecast - The report anticipates that the theoretical annual electricity consumption of computing centers will reach 600 billion kWh by 2030, accounting for approximately 5%-6% of the total electricity consumption in society [5] Industry Scale Growth Forecast - The computing power industry chain encompasses various segments, including chip R&D, server manufacturing, data center construction, and software systems. By 2030, the direct industry scale related to AI computing power is expected to reach around 2 trillion yuan, with the AI application industry scale projected to be about 5 trillion yuan, potentially driving an overall industry scale exceeding 15-20 trillion yuan [6] Recommendations for Statistical Standards - The report suggests establishing a unified statistical framework for computing power at the national level, categorizing it into intelligent computing, general computing, and storage computing. This would ensure consistency in reporting and avoid mixing traditional storage with intelligent computing totals [7] - It also recommends creating a unified statistical indicator system requiring computing centers to report both FP16 and FP8 computing power and actual energy consumption metrics for comparability across regions and types [7] - Additionally, a national-level data platform for computing power and energy consumption should be established to regularly publish authoritative data, aiding policy formulation and industry regulation [8]