Who can calculate the gross profit margin of coding packages for these AI big model companies? Anthropic's gross profit is approximately 40% The domestic models of Zhipu and Minimax are open-source, and those with computing power can also deploy them themselves. So judging that the coding package is not destined to sell too expensive, the gross profit margin may be extremely low, or even at a loss. Moreover, the business model of selling APIs is different from that of SaaS and other Internet applications. Each toekn has a cost, and there is no marginal cost close to zero. Zhipu still lacks computing power on its own. It sold too many cheap tokens earlier, and now the computing power package cannot continue to be sold. It is limited daily. Under the same business model, Anthropic does not have a limit, and it is unclear whether it is due to excessive spending or insufficient computing power that cannot be solved by adding more money. The downside of going public is that when the next financial report is released, we will know if they are swimming naked. The cost-effectiveness of Chinese models is hard to say, but it's hard to say how much profit the current coding package has and whether the business model is sustainable. Currently, it's more of a totem of faith, which is a very important investment target for AI storytelling. There's another interesting logic that I just figured out recently, 1. After the big model became popular, the first thing to benefit was the chip, which required computing power to train. Nvidia skyrocketed, and big model companies all had demand, competing with each other for the highest price to get the most premium and profit. 2. Moving on to memory, as the training and inference of AI large models require a significant amount of memory, the demand for inference continues to grow as the number of applications of large models increases. At this point, it is time to grab memory for inference, which is also the highest priced option, with room for premium. 3. Optical communication data transmission and explosive growth of AI computing power have made traditional supporting upgrades from copper to optical transmission the core link of AI computing infrastructure. Which field may be the next step? Companies that can provide computing infrastructure and services should be one of them. With the surge in demand for inference and insufficient computing power, increasing prices to buy computing power can provide more premium space. Amazon and Alibaba Cloud may have the opportunity to enjoy this premium.