Core Insights - The article highlights a significant shift in the AI industry from training to inference applications, indicating a growing demand for inference chips and related technologies [5][6]. Group 1: Chip Market Dynamics - The rental prices for leading inference chips have recently increased, indicating a rising demand in the market [3]. - In contrast, the rental prices for the main training chip, H100, have seen a decline, suggesting a shift in focus towards inference capabilities [4]. - The demand for AI model invocation has surged, with popular models like Claude 3.5/3.7 and Gemini 2.5 leading the charge [4]. Group 2: Usage Metrics - Microsoft Cloud reported a fivefold increase in token invocation from January to April, surpassing 500 trillion tokens in a single month [5]. - Alibaba Cloud also experienced a significant increase, with daily token invocation exceeding 30 trillion, marking a fourfold growth [5]. Group 3: Market Performance - The article notes that Broadcom, recognized as the king of ASIC chips, has outperformed Nvidia, the leader in training chips, reaching new highs [6]. - The AI application sector in the US stock market has also reached new highs across various industries, including military and education [6].