“钱花不出去！” AI故事被忽视的风险，正急剧升温

Group 1 - The core narrative is that the story of AI is shifting from "software eating the world" to "hardware being constrained by the world," highlighting a growing risk in the investment landscape due to political and physical limitations on data center expansion [1][4][15] - A rare bipartisan consensus has emerged between Senator Bernie Sanders and Governor Ron DeSantis regarding the need to halt the rapid increase of data centers, driven by public concerns over the negative impacts of AI [2][4] - The political landscape is changing, with states like New York, Arizona, Georgia, and Texas considering legislation to pause new data center projects or eliminate tax incentives, reflecting a growing backlash against the expansion of AI infrastructure [2][4] Group 2 - The anticipated capital expenditure of approximately $600 billion in 2026 faces skepticism regarding its feasibility, as major tech companies plan to spend $670 billion on AI infrastructure this year alone [4][6] - The energy demand from data centers is projected to double by 2035, raising concerns about the current U.S. electrical grid's ability to meet this demand, which could hinder the construction of new data centers [8][9] - The Texas power grid operator ERCOT is implementing a review process for power consumption projects, causing delays and uncertainty for tech companies, which could jeopardize their expansion plans and the associated capital expenditures [9][10] Group 3 - The financial markets are reacting to the risk of unspent capital, leading to significant sell-offs in tech stocks, as investors reassess the viability of AI-related investments in light of potential physical constraints [10][12] - The shift in market sentiment has resulted in a "de-leveraging" trend, with funds moving from high-beta tech stocks to more defensive sectors like chemicals and regional banks, indicating a growing fear of an AI valuation bubble bursting [12][13] - The market faces a dilemma: either trust that the electrical grid can expand to accommodate the projected $600 billion in capital expenditures or acknowledge that physical limitations have been reached, which would have severe implications for AI demand and investment [13][15]