Today's essay is about Meta starting to sell idle computing power. Bloomberg used the term "Sell Excess AI Compute", which is a very heavy term. It immediately leads people to wonder if Meta, the leading large enterprise, had bought too much computing power before, and speculated that semiconductor spending was excessive, leading to a sharp decline in the market before the market opened. Basically, this kind of ghost story comes every six months. The fermentation environment for this ghost story is that semiconductors have indeed increased significantly, and today they are even starting to speculate on the details of upstream components, indicating that the price has risen too much; The second is the gap period for upgrading Anthropic, OpenAI, and Gemini models. Since Opus 4.6 and Lobster, the market has not given stunning model performance. Once the models start to exceed expectations, people immediately worry that all the money spent will be in vain. But I don't think the music will stop now, for a simple reason: -Meta, like xAI, is struggling with their own models and initially felt that buying computing power was a pure loss without any inference income. However, whether the rental computing power was taken over ultimately depends on the token consumption and Capex trained; There is currently no clear data to prove that these two points will slow down; However, the short-term rocket like rise in the semiconductor sector will temporarily come to a halt. As I mentioned this afternoon, if major companies do not make profits and their stock prices do not rise, who will foot the bill for capital expenditures? Perhaps the market has given an opportunity to cut high and low