美股AI投资到底有没有泡沫

Core Viewpoint - The article emphasizes the need to abandon "bubble anxiety" and "scale worship," advocating for a long-term perspective in core technology investment and a pragmatic approach to commercialization for the healthy development of the AI industry [2][15]. Group 1: Structural Bubble - The debate over the AI bubble in the U.S. is fundamentally about the imbalance between high investment and low returns, manifesting differently across hardware, software, and application dimensions [16]. - In the hardware sector, the "computing power arms race" has led to uncontrolled capital expenditure, with NVIDIA being the primary beneficiary, showing signs of bubble pressure despite a 210% year-on-year increase in AI chip revenue for Q3 2025 and a gross margin of 78% [16][18]. - NVIDIA's stock price and valuation are increasingly characterized by bubble traits, with a current P/E ratio exceeding 75 times, significantly higher than the semiconductor industry's average of 30 times, and a market cap that once surpassed $3 trillion [16][18]. Group 2: Risks in the Ecosystem - NVIDIA's "binding prosperity" with the AI ecosystem poses a risk, as major clients like Microsoft and Google prepay large orders, creating a cycle that ties NVIDIA's performance to the financing heat of the AI industry [17]. - A 32% year-on-year decline in global AI startup financing in 2025 has led to some small clients canceling or delaying chip orders, resulting in a 15% quarter-on-quarter decline in NVIDIA's AI chip shipment growth [17]. Group 3: Capital Expenditure Trends - Major tech giants, including Microsoft, Amazon, and Google, are expected to exceed $470 billion in capital expenditure by 2026, doubling from 2024, with nearly 60% directed towards NVIDIA, amplifying the risk of over-investment [18]. - Oracle's capital expenditure for FY 2026 has been raised to $50 billion, a 136% increase year-on-year, which constitutes 75% of its revenue, leading to a negative free cash flow of $10 billion [18]. Group 4: Software Sector Challenges - The software sector is experiencing a commercial shortfall masked by circular financing, with OpenAI planning to invest $1.4 trillion over several years but still projected to incur a loss of $115 billion by 2029 [19]. - The valuation of leading AI companies is severely disconnected from their performance, with Palantir's P/E ratio exceeding 180 times and Snowflake nearing 140 times, raising concerns about the sustainability of these valuations [19]. Group 5: Application Bottlenecks - The commercialization bottleneck is increasingly evident, with few scalable profit-generating scenarios despite the unprecedented popularity of generative AI [20]. - Major tech companies' AI-related revenue growth is insufficient to cover their substantial capital expenditures, leading to negative free cash flow projections for companies like Meta and Microsoft by 2026 [20]. Group 6: Comparative Analysis of U.S. and China - China's AI investment is characterized by "excessive rationality and insufficient heat," with a total capital expenditure of approximately 400 billion yuan by 2025, only one-tenth of that of U.S. peers [22]. - Chinese companies are avoiding the U.S. path of "stacking computing power," making steady progress in domestic chip replacement, while local AI models are rapidly iterating and adapting to domestic application scenarios [23]. Group 7: Strategic Differences - The differences in AI investment strategies between the U.S. and China stem from their respective development models, with the U.S. adopting an aggressive approach and China focusing on steady progress while controlling risks [24]. - For the U.S., addressing AI bubble risks involves shifting investment focus from computing power accumulation to technological innovation and efficiency improvement [24].