申万宏源冯晓宇：美股AI行情进入“换挡期”

Group 1 - The core concern in the market regarding the "AI bubble" is the gap between the time it takes for new technologies to generate economic scale effects and the optimistic expectations of capital market returns from these technologies [2][3] - Since 2023, the current AI market in the US has been ongoing for three years, primarily focusing on valuation increases, with widespread growth in AI infrastructure and profitable AI applications [1][3] - By 2025, the contribution of valuations from AI-related industries (technology, communication) is expected to be limited, with increased reliance on debt financing for investments, leading to a more stringent market evaluation of returns on investment (ROI) in the AI sector [1][5] Group 2 - Hard constraints affecting the AI sector often lead to periodic adjustments in stock prices, with the "Buy the Dip" strategy still showing effectiveness in the US market [3] - Since 2023, each round of adjustment in the AI sector has been triggered by macro liquidity tightening and concerns over computing power, algorithms, and electricity [3][4] - The soft constraints, when they manifest, could trigger a systemic bubble burst, necessitating a strategic reduction in positions based on ROI indicators [4][5] Group 3 - The cost of inference for AI models has been decreasing, while the training costs for cutting-edge AI models are rising significantly, indicating a divergence in cost dynamics [4] - The penetration rate of AI in US enterprises is currently around 10%, with higher rates in information-intensive sectors, and is expected to rise in the next six months [4][5] - The total investment commitment for AI by foreign and US companies is projected to be approximately $3.8 trillion by 2028, with ROI sensitivity to GPU depreciation periods being a critical factor [4][5] Group 4 - Under a neutral assumption, the risk of soft constraints in AI by 2026 is manageable, with a focus on the penetration rate of B-end enterprises and the stability of cash flows [5] - In an optimistic scenario, the acceleration of B-end enterprise penetration and high growth in AI infrastructure capital expenditures (capex) could lead to increased profitability for cloud companies [5] - In a pessimistic scenario, inflation pressures could resurface in the second half of 2026, potentially leading to valuation corrections if AI application costs do not decrease sufficiently [5]