Core Insights - The report from Guosen Securities highlights that advertising and SaaS products are high-demand areas based on token consumption and input-output analysis [1] - There is a significant need for continued support from downstream market conditions for the massive CAPEX investments, particularly in sectors like autonomous driving and robotics, which are driving GPU leasing demand [1] - The report indicates that without a trend of explosive growth in downstream applications, leading companies may slow their CAPEX spending, especially those reliant on single customers [1] Supply Side - The financial pressure from high CAPEX growth and power shortages is evident, with major overseas companies' CAPEX exceeding 60% of their operating cash flow [2] - Companies are increasingly resorting to financing leases and financial operations to manage cash flow constraints, with firms like Meta and Google utilizing debt financing to supplement their investments [2] - The projected growth in data center capacity in North America is expected to be 100% in 2026 and 50% in 2027, leading to significant power shortages [2] Demand Side - There remains a substantial gap between the cumulative incremental CAPEX of overseas cloud providers since the end of 2022 and the current backlog of unfulfilled orders, which is less than 50% of the current order scale [3] - Excluding Oracle, which is heavily impacted by OpenAI, the three major cloud providers are still in a state of supply shortage, indicating that current CAPEX levels are not excessive [3] Computing Power Demand Structure - The current structure of computing power demand shows a higher internal proportion for cloud providers, supporting AI applications, model training, and the development of AI products like Gemini and Copilot [4] - External revenue is also growing rapidly, primarily from GPU leasing and API calls for AI application development, with GPU leasing accounting for about 70% of demand [4] - API calls, while currently a smaller portion of demand, are growing quickly, with many small companies utilizing models for customer service and process optimization [4]