Workflow
大厂“AI烧钱大战”:当下规模被低估,未来折旧被低估,最早2027年爆发价格战
Hua Er Jie Jian Wen·2025-09-18 09:18

Core Insights - The current AI infrastructure investment by major tech companies is unprecedented and approaching the peak levels seen during the internet bubble [1] - The market is significantly underestimating the scale of current AI investments and the future depreciation costs associated with these investments [1][9] - A potential supply-demand imbalance in cloud services could lead to a price war as early as 2027 if supply continues to outpace demand [1][14] Investment Scale Underestimation - Morgan Stanley's report indicates that capital expenditure (Capex) by "super-scale" players like Amazon, Google, Meta, Microsoft, and Oracle is projected to reach 26% of sales by 2027, nearing the 32% peak during the internet bubble [1][2] - The use of off-balance-sheet financing tools, such as leasing, is increasingly common, leading to an underestimation of actual investment levels [1][5] Factors Contributing to Underestimation - The rise of financing leases allows companies like Microsoft and Oracle to build data centers without fully reflecting these costs in traditional Capex figures, significantly increasing their capital intensity [5] - The "Construction in Progress" (CIP) assets are accumulating on balance sheets without being depreciated, meaning the financial impact on profits has yet to be realized [7] Future Depreciation Costs - Analysts at Bank of America highlight that Wall Street is slow to react to the anticipated increase in depreciation expenses, with a projected shortfall of nearly $16.4 billion in depreciation costs for Google, Amazon, and Meta by 2027 [9] - The expected depreciation for these companies is significantly underestimated, with Alphabet facing a gap of approximately $7 billion, Amazon $5.9 billion, and Meta $3.5 billion [9] Short Lifespan of AI Assets - AI-related hardware, such as GPUs, has a shorter effective lifespan of three to five years due to rapid technological advancements, which could accelerate depreciation costs [13] - Amazon has already reduced the expected lifespan of some servers from six years to five years, indicating a shift in asset management strategies [13] Potential Price War - There is a risk of overcapacity in the AI infrastructure market, which could lead to aggressive pricing strategies by major tech firms if supply exceeds demand [14] - The increasing similarity in performance among large language models may further commoditize infrastructure services, prompting companies to adopt more aggressive pricing to maintain utilization rates [14]