据中国网消息，9月17日上午，国务院新闻办公室举行"高质量完成'十四五'规划"系列主题新闻发布会。会上，国务院国资委副主任李镇介绍道，"十四 五"期间，智算供给能力显著提升，通信运营商加快向AI基础底座供应商转变，累计投资超百亿元，建成4个万卡集群。 这四个万卡集群分别为中国电信京津冀智能算力中心、中国电信临港智算中心、中国移动智算中心（哈尔滨）以及中国移动智算中心（呼和浩特）。 据介绍，万卡集群，是指规模数量由一万张及以上的加速卡（包括GPU、TPU及其他专用AI加速芯片）组成的高性能计算系统，用于加速人工智能模型的 训练和推理过程。由于AI大模型的训练极度消耗底层算力，因此，通过集成巨量规模的单卡算力芯片，成为当前算力基础设施部署的主流趋势。 在国内，除了电信、移动为代表的运营商外，包括字节跳动、阿里、百度、华为及科大讯飞等在内的互联网和AI头部企业均在发力万卡集群的建设。 中国移动智算中心（哈尔滨） 图片转自智通财经 五是开源生态不断完善。发起设立了"焕新社区"，面向广大青年创业者、科技爱好者、高校院所等，免费提供国产智算芯片超过2000张，汇聚了244个垂 类行业模型、近160个高质量数据集，有效助力 ...

Summary of Key Points from the Conference Call Industry Overview - The conference call discusses the **computing power leasing industry**, highlighting its attractiveness due to the combination of AI and stablecoins, leading to significant market expectation discrepancies [1][3][11]. Core Insights and Arguments - **Profitability of WanKa Clusters**: Ordinary WanKa clusters are estimated to generate an annual profit of **150-200 million RMB**, while advanced clusters can reach **300-400 million RMB**, significantly enhancing the price-to-earnings ratio of listed companies [1][5][6]. - **Shift from CAPEX to OPEX**: Domestic internet companies are transitioning from capital expenditures (CAPEX) to operational expenditures (OPEX), which is driving the growth of the computing power leasing market as they prefer leasing over purchasing high-performance computing equipment [1][8][11]. - **Five-Year Contract Announcements**: Companies are announcing five-year contracts to ensure stable future revenues. For instance, Hainan Huatie announced a **3.7 billion RMB** contract, which is expected to generate substantial profits over five years [1][13]. - **Revenue Sources**: The main revenue sources for five-year computing power leasing contracts include leasing income (approximately **4 billion RMB** for ordinary cards and **6 billion RMB** for advanced cards), tax refunds (around **200 million RMB**), and residual value income (about **130 million RMB**) [1][14][16]. - **Stablecoin Applications**: The application of stablecoins in the computing power leasing industry is promising, as it aligns with predictable cash flow businesses. Companies may issue RWA tokens to tokenize mining profits and expand financial leverage [1][22]. Growth Potential - **Market Size and Growth**: China's AI capital expenditure is expected to reach **1 trillion RMB** by 2028, with **50%** allocated to computing power leasing, indicating a market size of approximately **400-500 billion RMB** and significant growth potential [2][11]. - **Investment Value**: The computing power leasing industry is projected to have a price-to-earnings ratio of **5 to 10 times** by 2026, with high certainty in performance growth, particularly due to the integration of AI and stablecoins [3][11][31]. Competitive Landscape - **Limited Competitors**: Only a few companies in China can achieve WanKa cluster scale, creating high technical barriers and competitive advantages, especially those closely associated with stablecoin businesses [4][9]. - **International Insights**: The success of overseas companies like Cowife, which transitioned from cryptocurrency mining to AI infrastructure, serves as a model for domestic companies [7]. Financial Performance - **Revenue and Profitability**: A typical company in the industry reported revenues of **1.9 billion USD** last year, with a projected annual revenue of **6-7 billion USD** in the coming years, assuming a net profit margin of **20-30%** [10][12]. - **Cost Structure**: The costs for five-year computing power leasing primarily include server costs, financing costs, and operational maintenance costs, with a significant portion funded by financial institutions [17][19]. Future Outlook - **Profit Projections**: Companies planning multiple advanced WanKa clusters can expect substantial profits, with estimates suggesting a total profit of **2 billion RMB** from five clusters, alongside additional profits from core operations [18][26]. - **Market Trends**: The computing power leasing market is expected to thrive, driven by the increasing demand from large domestic companies transitioning to OPEX and the overall growth of the AI sector [30][31].