Tai Mei Ti A P P·2025-11-03 05:58Core Viewpoint - The value in the AI infrastructure market is expected to shift from chip manufacturers like NVIDIA to cloud service providers, but this perspective is challenged by the current economic realities of data centers and intense competition from Chinese firms [1][2]. Group 1: Cloud Profitability Reality - Cloud service profitability is significantly lower than anticipated, as evidenced by Oracle's data center gross margin disclosures, which reveal limited profitability even after accounting for GPU depreciation [3][4]. - The economic model resembles a monopoly upstream (NVIDIA) extracting major profits, while the downstream (cloud services) faces fierce competition and high costs, leading to limited profits [4][5]. Group 2: Downstream Application and Competition - Downstream profits are not as optimistic as market expectations, with many enterprises finding token demand lower than anticipated and economic benefits from tokens not materializing immediately [6][7][8]. - The competition from Chinese firms, particularly Alibaba's Qwen series, is reshaping cost structures with significantly lower token prices, creating downward pressure on pricing [10][11][20]. Group 3: Capital Cycle Perspective - The AI infrastructure investment is projected to exceed $4 trillion over the next 5-7 years, with signs of overcapacity and declining unit profits reminiscent of the historical railway boom [23]. - Key indicators of a potential cycle turning point include declining utilization rates, intensified price wars, and tightening financing conditions [23]. Group 4: Future Value Distribution - The ultimate value in AI is expected to flow towards application developers rather than cloud service providers, as chip manufacturers currently extract high profits, but as model costs decrease, the real profits will shift to AI enterprise software and vertical industry solutions [24][25]. - Alibaba's low-price strategy for tokens indicates a future where tokens become cheap and replaceable resources rather than high-margin products [26].