Serenity: Amazon is the clearest investment logic for supercomputing AI, but the risk of supercomputing profits being eroded by semiconductors or triggering a turning point in capital expenditures According to BlockBeats, on June 24th, market analysts recently discussed the sustainability of AI capital expenditures by supercomputing vendors, with clear differences in opinions. Serenity believes that Amazon is currently the most convincing investment target for supercomputing. The logic is that Amazon has a huge human resource base of approximately 1.57 million employees, and AI capital expenditures can be directly converted into internal cost reduction - by automating labor through big language models, optimizing delivery through autonomous driving, compressing operating costs through warehouse robots, and adding AWS computing power expansion (self-developed Trainium chips) and potential chip sales revenue, the business path is clear. Google is ranked second, with AI investment supported by multiple factors such as defending search moats, Google Cloud revenue, and advertising optimization, but the physical AI layout is relatively weak. Microsoft and Meta are considered to have not yet explained the necessity of capital expenditures to the market, with Microsoft's Maia chip lagging behind and AI strategy constrained by OpenAI investment, resulting in low market sentiment. However, doubts cannot be ignored. Some viewpoints point out that the ultimate beneficiaries of large-scale capital expenditures by current supercomputer manufacturers are largely semiconductor companies represented by Nvidia, and the profit margins of supercomputers themselves are continuously compressed. Once supercomputer manufacturers lose the motivation to expand their spending, or downstream model companies such as Anthropic experience slower than expected revenue growth, the current market valuation system based on AI capital expenditure narrative will face the risk of sudden reassessment. Serenity emphasizes that various supercomputer manufacturers are currently seizing the "leading big model effect" - platforms with the strongest models will continue to accumulate training data and expand their capability advantages. This structural gap may be solidifying, which is also a deep driving force for all players to accelerate construction. [Original link]