The overall thinking on AI stocks in the US stock market and speculation about future economic cycles triggered by the decline of Oracle - Yesterday, the US stock market fell sharply, with Broadcom falling more than 11% and Oracle falling more than 4%. Although it cannot be said to be Black Friday, it is also a tough day for Broadcom and Oracle. The heavy losses of Broadcom and Oracle directly dragged down the tech giant sector, and other chip stocks such as Nvidia also generally followed suit. In contrast, the Dow Jones Industrial Average has performed relatively well due to the support of traditional value stocks, and the market shows clear signs of "abandoning technology and moving towards value". Oracle's sharp decline this time is due to the announcement that it will directly increase its capital expenditure budget for the 2026 fiscal year from $35 billion to $50 billion. The market is concerned that the bottomless pit of burning money to build data centers will continue to result in negative cash flow, and the company is at risk of losing blood. The situation of Broadcom is completely different from Oracle. It was mistakenly killed. In fact, Broadcom's financial report is very beautiful, but at this sensitive time, the market is using a magnifying glass to find faults. The reason is that the proportion of customized AI chips (ASICs) is increasing, and the gross profit of such businesses is usually lower than that of Broadcom's traditional general-purpose chips. The market is concerned that as the proportion of AI business expands, the overall profit margin will be lowered. After excluding the impact of mergers and acquisitions, the gross profit margin was 76.6%, slightly lower than the previous period. In fact, Oracle has turned itself into the "lever amplifier" of the AI foam through the leveraged gambling data center. If AI succeeds, it is the big winner; But if AI is just a common prosperity rather than a "singularity", Oracle's model of short-term debt and long-term investment, heavy assets and high leverage is highly likely to lead to a serious liquidity crisis. However, labor shortages and power bottlenecks are highly likely to cause project delays. Oracle's investment grade bonds are being traded as quasi junk bonds, and its continuously rising CDS (credit default swaps) costs have made the market smell the "subprime crisis". This indicates that the market is beginning to doubt not short-term profits, but the ability to recover capital over a period of ten to twenty years. When the credit side begins to question, the high valuation of the equity side will inevitably be compressed, and this compression is usually fast and fierce. By comparison, Broadcom's decline was only due to the fact that "telling the truth" shattered the illusion of high profits. That's also why Broadcom fell 10% (just worried about profit margins), while Oracle is seen by the market as a "systemic risk". The plunge on December 12 was a necessary foam squeeze. The market is transitioning from blindly speculating on infrastructure (buying shovels) to forcibly assessing investment returns (calculating), which continues the logic of Nvidia's previous encounter with Google Gemini3+TPU attacks. From the perspective of the large economic cycle, AI is more like the early stage of a new round of general technology revolution than an industry that has entered a mature cashing period, which means that its role in the macro cycle is closer to the early form of railways, electricity and the Internet, that is, it first experiences long-term high-intensity investment, low returns and even negative cash flow, and then cashes in productivity dividends through efficiency leaps and application diffusion in the later stage. Therefore, in the mid-term economic fluctuations, AI often exhibits the characteristics of "pro cyclical valuation and countercyclical profitability". When the economy and liquidity are good, the market is willing to pay for long-term imagination, and AI related asset valuations expand the fastest; Once entering the policy turning point or capital discipline stage, the assets with the longest duration and the most sensitive to the future are also the first to be corrected. (Same goes for Bitcoin) This does not mean that AI will lose its core position in future cycles. On the contrary, the real law is that AI will continuously transform from a "narrative asset" to "infrastructure" and "production factors" in every round of economic downturn or adjustment. After valuation is compressed and capital expenditure is constrained, surviving enterprises are more likely to become the core carrier of efficiency improvement in the next round of recovery. In other words, the short-term sharp decline reflects the market's brake on the overly fast, heavy, and concentrated pace of AI investment, rather than a denial of the AI technology roadmap itself; In longer economic cycles, AI will not simply follow the ups and downs of the economy like cyclical industries, but will present a wave like evolution of multiple rounds of "valuation first, reality lagging, and re evaluation".