数据不会说谎。2025年全球AI芯片市场规模突破1200亿美元，台积电、三星的3nm产线火力全开，单块 GPU的算力三年翻了10倍。可另一边，全球电力增速常年维持在3%左右，欧洲甚至因能源危机砍了数 据中心的供电配额。马斯克的焦虑不是空穴来风：当单个AI集群的功耗突破100兆瓦（相当于一座小型 核电站），当数据中心耗电量占全球10%且还在飙升，地面上的"电"确实快不够用了。 "记住我的话，36个月内太空会成为部署人工智能的最便宜去处。"马斯克的语气带着惯有的笃定，像极 了当年宣布"殖民火星"时的决绝。但这次，他的论据扎得很痛："芯片产出几乎呈指数级增长，但电力 产出是平的。那么你要如何让这些芯片通电？靠魔法电源？魔法电力精灵吗？" 当马斯克用"魔法电力精灵"调侃地面电力困境时，黄仁勋正拿着散热板图纸算尺寸。一个在太空画饼， 一个在现实算账——这场关于"太空数据中心"的隔空对话，撕开了AI时代最尖锐的矛盾：人类对算力的 贪婪，正在把地球逼到电力极限。 二、黄仁勋的"冰冷现实"：没空气的太空，散热板得造到"看得见" 就在马斯克畅想太空算力乌托邦时，黄仁勋在财报电话会议上泼了盆冰水："在太空中，能源与散热是 两回事 ...

Summary of Key Points from the Conference Call Industry Overview - The conference focuses on the AI computing power sector, highlighting the significant growth driven by inference and training demands, emphasizing the importance of large models and applications rather than solely relying on industry chain data [1][2][3]. Core Insights and Arguments - **AI Computing Demand**: The demand for computing power is closely linked to the volume of tokens, with increasing computational needs driving this trend. The growth in overseas computing companies cannot be explained solely by traditional performance metrics, necessitating a deeper analysis of how token volume influences computing demand and future trends [1][4]. - **Agentic AI Concept**: Introduced as a new paradigm derived from reasoning models, agentic AI emphasizes task distribution, execution, and planning to achieve specific goals, capable of handling complex or simple tasks through a multi-step process [1][6]. - **GTC Conference Attendance**: The GTC conference saw a 50% increase in attendance compared to the previous year, with a notable rise in AI industry participants, indicating the growing importance of the event for the AI sector [3]. - **Token Explosion**: The global token volume is experiencing explosive growth, significantly impacting computing demand. The relationship between token consumption and computing power is complex and non-linear, with a potential for exponential increases in demand [12][17][21]. Important but Overlooked Content - **Skin Law**: Huang Renxun introduced the concept of "skin law," which describes the inflation of computing demand across three phases: pre-training, post-training, and test time, each contributing to increased computational needs [8][10]. - **Future Drivers of Computing Demand**: The shift from CPU to GPU architectures and the need for capital investment in software rather than just human resources are identified as key factors driving future computing demand [34][35]. - **Market Dynamics**: The competition among major tech companies to enhance user experience through faster response times and accurate outputs is leading to increased investments in computing power, indicating a shift towards a model where software relies heavily on computational resources [26][38]. Market Predictions - **Data Center Market Growth**: The data center market is expected to exceed $1 trillion by 2028, with 2025 being a pivotal year for rapid growth in demand [32]. - **GPU Demand**: Major cloud service providers have shown significant demand for GPUs, purchasing millions of units, indicating sustained growth in computing needs [31]. Conclusion - The AI computing power sector is at a critical juncture, with emerging paradigms like agentic AI and the explosive growth of token consumption reshaping the landscape. Understanding these dynamics is essential for accurately predicting future trends and making informed investment decisions in the sector [5][43][45].