$530B in AI Capex looks terrible if you forget how accounting works

The article discusses the massive capital expenditure (capex) of $530 billion by major tech companies like Microsoft, Amazon, and Alphabet (Google) on AI infrastructure such as GPUs, servers, and data centers. The author argues that the market's negative reaction to these capex figures is misguided, as it fails to account for how accounting principles work. The key points are: 1. The depreciation of these assets over 5-6 years for servers/GPUs and 15-25 years for buildings/infrastructure creates a significant non-cash expense that depresses reported earnings, even though the actual cash flow generated is much higher. 2. Adjusting for the replacement of fully depreciated equipment and lower incremental operating expenses, the cumulative net income over the asset's useful life is estimated to be around $392 billion, or an 11% annualized return. 3. However, the author argues that the more relevant metric is the free cash flow (FCF) yield, which is estimated to be around 27% annually, with the capex cohort paying for itself in 3.5 years. 4. The author suggests that the market is incorrectly evaluating these infrastructure investments based on accounting earnings rather than cash returns, which is a category error given the nature of these capital-intensive assets. 5. The author acknowledges the bear case around accelerated hardware obsolescence and disappointing demand, but argues that the current growth rates of the cloud businesses suggest the capex is justified. Overall, the article presents a nuanced perspective on how to properly evaluate the massive capex investments by tech giants in AI infrastructure, emphasizing the importance of considering cash flow metrics rather than just accounting earnings.