Some thoughts:
1. In this round of bull market, I see many people guessing the top. In fact, 4000, 5000, and 6000 are meaningless, and the key is to look at the real estate market. Every past bull market has been accompanied by a jump in housing prices, a large transfer of funds, and a drain on funds. If this round of the real estate market is still like this, then the bull market will have no top, reshaping a generation's asset beliefs. Conversely, if not, then run quickly, history is still repeating itself.
2. Trump has indeed done a great job this time. The EU, Japan, and South Korea have all kneeled, and funds have flowed back to the United States in large amounts, which is good for Nasdaq and AI infrastructure. In fact, to analyze various assets, you have to look at the flow of money.
3. Anti-involution must be combined with demand-side policies. The success of the past supply-side was also inseparable from the demand-side. For example, the beer industry now has no involution on the supply-side, but it still doesn't work, which is the impact of deflation.
4. Will the fertility subsidy be "increased layer by layer" in the future? It will be good to have babies in the future. I even thought about whether having children in the future will be like technology subsidies now. For example, many so-called technology companies don't make any money, but the state subsidizes, the province subsidizes, and various departments in the city subsidize, and they get subsidies to the point of softness, resulting in overcapacity. If the zc can really shift from the supply-side to the demand-side, then this phenomenon is likely to occur.
5. Still that sentence, the 15th Five-Year Plan determines the direction of 💰, and the analysis of various assets is inseparable from this.
AI aspect:
6. The "underperformance" of GPT5 was actually disclosed by information five days ago (I even think this is a message deliberately released by OpenAI in advance to manage expectations). The real reason behind this is that Silicon Valley now has a new consensus: that is, no longer pursue the cross-cutting function of the model itself, but pursue the practicality of the model (Oai already has 700 million users worldwide, no longer a research institution pursuing AGI, but DS chose the opposite route). Now Silicon Valley and Wall Street's evaluation of AI has turned to the "Economic Turing Test", that is, when an AI completes a task, you can't tell whether it is a person or a machine behind it. In other words, as long as AI can effectively improve productivity (whether you are AGI or not), this model is successful.
7. Following the above, when your users are at the level of 1 billion, the importance of practicality is greatly improved, because even if the production efficiency is only improved by one-thousandth, the increase in GDP is a terrifying number. Therefore, OpenAI is not unable to create the kind of cutting-edge world model recently released by Google that makes ordinary people "wow", but it has strategic trade-offs. Therefore, the "underperformance" of GPT5 was already within the expectations of Wall Street, and US AI hardware stocks have been soaring these days.
8. The US AI capex expenditure is expected to account for 25% of the actual growth of the US GDP in 2025. The infrastructure freak is well-deserved. I have always said that Da Meili has been the No. 1 infrastructure country in history (historically, railway capex accounted for 6% of the total GDP),
![[狗头]](https://public.bnbstatic.com/image/pgc/202508/2e6cf1bfaefacf65f04af19fda76b7e4.png)
), but in recent decades, I haven't found the direction. We are at most the No. 2 infrastructure boss. But don't worry, we won't be absent from this kind of thing.
.
9. Now GPT, Gemini, and Claude are the main AI applications on the other side, with a total WAU of 1 billion. And the total of all our AI applications is less than one-tenth of theirs, so in terms of AI, we are already two species, just like we go abroad to see other people's mobile Internet as if we are watching primitive people.
10. From Meta's actions, we can see that it is people and cards, or algorithms and computing power in a nice way. I now use this very simple standard to judge whether a certain target is worth betting on, whether you are making models, making applications, or making ecosystems. Many companies in our A-shares are covered with the concept of AI, but they have neither talent nor cards (in fact, people are scarcer than cards), so what can they use to undertake the value of AI, just pass it, and you are unlikely to miss it.
11. GPT5 used synthetic data and a new post-training paradigm, so the data barrier isn't as high as many people think. In fact, after so many years of talking about big data, the data barrier has always belonged to large companies. I have never seen a small company truly use data as a moat.
12. From chips and tariffs, we can see that our opponents are also growing, and their methods are becoming more professional and mature. Alas, I sincerely hope that we can achieve internal breakthroughs.
13. Most domestic primary market VCs are still betting on robots, and some are betting on AI hardware. Very few are betting on models and applications. This is just a phenomenon, and everyone can analyze it for themselves.