美股IPO·2025-09-18 11:53Core Viewpoint - Major tech companies are entering an unprecedented AI infrastructure arms race, with capital expenditure intensity nearing the peak levels seen during the internet bubble, indicating a potential underestimation of current AI investment and future depreciation costs [3][4][11] Group 1: Capital Expenditure Trends - Morgan Stanley predicts that by 2027, capital expenditure as a percentage of revenue for major players like Amazon, Google, Meta, Microsoft, and Oracle will reach 26%, close to the 32% peak during the internet bubble and exceeding the 20% during the shale oil boom [3][4] - The actual scale of investment is underestimated due to the increasing use of off-balance-sheet tools like financing leases, which allow companies to accelerate data center expansion without fully reflecting these investments in traditional capital expenditure data [5][7] - Microsoft and Oracle's capital intensity is expected to rise significantly when financing leases are accounted for, with Microsoft's capital expenditure to sales ratio projected to jump from 28% to 38% by FY2026, and Oracle's from 41% to 58% [7] Group 2: Impact of Construction in Progress - A significant amount of capital is currently tied up in "Construction in Progress" (CIP), which does not incur depreciation until the assets are operational, meaning the financial impact on profits has yet to be realized [9] - Google, Amazon, Meta, and Oracle have seen substantial increases in their CIP balances, with Amazon's growing by approximately 60% ($17 billion) and Google's by about 40% ($15 billion) over the past year [9] Group 3: Future Depreciation Costs - Analysts at Bank of America highlight that Wall Street is underestimating future depreciation costs, with significant discrepancies expected by 2027: $7 billion for Alphabet (Google), $5.9 billion for Amazon, and $3.5 billion for Meta, totaling nearly $16.4 billion in expected shortfall [11] - The rapid technological advancements in AI hardware, such as GPUs, may lead to shorter asset lifespans, with Amazon already reducing the expected lifespan of some servers from six years to five due to accelerated technology development [13] Group 4: Potential Market Risks - Bank of America warns that the AI infrastructure market may face a repeat of historical patterns where aggressive investment leads to overcapacity and pricing pressures, with the risk of a price war emerging as early as 2027 if supply outstrips demand [14] - Major tech companies are ramping up AI infrastructure investments, which could result in a scenario where the supply of computing power exceeds the demand for high-value AI services, potentially leading to aggressive pricing strategies to maintain utilization rates [14]