Shawn said a very good question: The biggest difference between the AI era and the Internet era is: 1) After the fixed costs (servers, bandwidth) are invested in the Internet/mobile Internet era, the marginal cost is almost zero, the more users → the stronger the network effect → the higher the valuation (the DAU/MAU penetration rate is the core KPI); 2) The reality of the AI era, especially generative AI, is the actual consumption of computing power charged by tokens. The free model is essentially a subsidy from big model companies for users to burn money and run models. The higher the user penetration rate, the larger the total amount of tokens processed per day, and the direct linear increase in graphics card/electricity/cloud vendor bills. This question accurately points out an unresolved "cost scale" contradiction in the AI business model. From a personal perspective, penetration rate remains an effective indicator, as it is a demand side indicator (whether it is a B-end or C-end indicator): in the early stages of any technological revolution evolution iteration, we must first look at the user side, and then look at the cost side. Several logics: 1) In the Internet era, the penetration rate (traffic) should be seized first, and then the realization should be discussed. AI is currently in an early explosive period similar to the "iPhone moment". The higher the user/enterprise adoption rate, the faster the ecosystem, network effects, and data flywheel will rotate, which is the primary driving force for industry growth; 2) Industrial growth logic: In the early stages, AI companies (especially AI native) still need to consider demand side indicators such as user growth, retention, DAU/MAU/WAU, and token usage. The core is that the user base has increased, and the absolute value of the paying population is larger under the same payment rate. At the same time, with a larger user base and better technical experience, the payment rate will also increase synchronously. With rapid user growth and reaching a turning point in commercialization, early investments can be recouped more quickly. 3) Many large companies have similar indicators Meta even treats token consumption as an internal KPI for AI penetration rate. The growth of the industry first depends on the user side, and penetration rate is still the hardest indicator on the demand side. Without users, there is no follow-up. Of course, Shawn's question also illustrates that in the new field of AI with high marginal costs, simply pursuing penetration while pursuing commercialization and expanding the paying population can easily become a trap of "bigger scale, more money burning". This is actually the second step in penetration rate, which is to increase conversion rate (payment rate) and increase the proportion of high ARP users on the basis of rapid growth in penetration rate. So as we move forward, we need to look not only at penetration rate, but also at "effective penetration rate (payment rate and high ARPU ratio): while users skyrocket, unit token cost, gross profit margin, and cash flow can also keep up. We can also see that AI big model manufacturers are actually moving faster towards commercialization, starting from the second year of their rise and seeing rapid growth in commercial ARR this year. In short: The first stage is to look at the increase in penetration rate, with better user acceptance and a larger base. The second stage depends on the monetization ability, which is the increase in payment rate/the proportion of high ARPU users, and of course, the consumption of tokens can also be directly observed. It was also discussed on June 3rd in 168X Space about the revenue growth of core large model manufacturers and the cloud business growth of large technology manufacturers. This is the most direct manifestation of monetization ability and commercialization. This article is sponsored by @ bitget_zh, titled 'Bitget Buying US Stocks: Instant Entry, Smooth Trading'