The largest AI foam in human history has arrived, and the collapse is near In 2026, the global capital investment in the AI field will be insane, with the four giants such as Microsoft and Google spending over $600 billion on AI in a single year. This scale far exceeds the industry's conventional research and development boundaries, making it the largest radical capital investment in human history. However, behind this carnival, there is a financial crisis similar to that of LeEco in those years, even a huge one, which has become the largest foam in human history and is not far from collapse. LeEco used the straight-line method to amortize the cost of "perishable products" such as film and television copyrights over a period of 10 years, putting profits first and costs later, in order to whitewash its financial reports. Eventually, it collapsed due to inflated assets and broken funding chains. Nowadays, AI giants are repeating this logic: Nvidia GPUs iterate at five times the speed of traditional Moore's Law, releasing disruptive new products in just one year. The previous generation of graphics cards quickly became electronic waste, and GPUs have become "perishable products" that should have accelerated depreciation within two years. However, major technology companies still use the traditional rule of straight-line amortization over six years, artificially lowering current costs and making financial reports appear impressive, but actually delaying huge depreciation risks to the future. What's even more frightening is that almost all AI products nowadays adopt a subscription system, perfectly replicating LeEco's accounting magic of "left foot subscription system and right foot cost shifting" in the past, creating false prosperity on the books. The purchasing of computing power by tech giants has long become a prisoner's dilemma defensive game: knowing that hardware will depreciate rapidly, they have to grit their teeth and borrow money to place orders crazily. If they don't follow up, they may be eliminated by the industry and ultimately forced into a vicious cycle of high investment and high debt. This scene is also highly similar to the foam of SciDev. Net in 2000: when enterprises bought servers crazily, rapid hardware iteration led to a sharp depreciation of assets, and supplier financing planted a financial bomb. Finally, the Federal Reserve raised interest rates, tightened liquidity, and a large number of companies went bankrupt. NASDAQ plummeted nearly 80%. Now, Nvidia has achieved the same level of AI computing power as Intel, Oracle and other companies did in the past, with expected capital expenditures soaring to $50 billion by 2026, net debt exceeding $100 billion, debt to equity ratio reaching 500%, and credit default risk soaring; Google issues 100 year bond renewal, burning billions of dollars every day on AI computing power; Microsoft has also invested in OpenAI through cloud service vouchers, forming a financial closed loop where the risks of startups are tied to their own balance sheets. The craziness of the market is often more lasting than rationality. No one can accurately predict the time when the foam will burst. Financial reports are less than expected, power cannot support the expansion of computing power, and bad debts cannot be covered up, all of which may become the last straw to crush the foam. But what is certain is that the prosperity that deviates from common sense will end, and the strength of foam bursting will be far beyond imagination. This post is reposted from Tang Shan Lao Wang AI foam macroeconomic investment risk market early warning