Bank of America Merrill Lynch: First gives "buy" ratings to KNOWLEDGE ATLAS (02513) and MINIMAX-W (00100), the global AI market is forming a dual-speed pattern.

Bank of America Corp has released a research report stating that it has given KNOWLEDGE ATLAS (02513) and MINIMAX-W (00100) a "buy" investment rating for the first time, with target prices of 1250 Hong Kong dollars and 500 Hong Kong dollars respectively. Bank of America stated that the global artificial intelligence (AI) market is forming a dual-speed pattern: U.S. companies may maintain a leading edge in cutting-edge capabilities, while Chinese models are gaining market recognition due to their low cost and high performance. Despite limited computing power, Chinese companies are optimizing expenditure strategies - as the prices of cutting-edge models rise, this cost-effective advantage is becoming increasingly important globally, especially in large-scale, non-critical application scenarios. The bank stated that the AI model market in China is vibrant and fiercely competitive, with limited differences in capabilities between existing companies (such as Alibaba Group Holding Limited Sponsored ADR(09988/BABA.US), ByteDance) and independent AI labs (such as DeepSeek, Zhipu, MINIMAX). In this environment, slight advantages in model efficiency may not bring lasting profits - computing resources, ecosystem integration, and distribution channels are key moats, which structurally favor existing companies. Bank of America indicated that with stronger balance sheets and integrated cloud platforms, existing companies are better positioned to withstand lower model layer profit margins, thereby driving cross-selling of computing resources. At the same time, the constantly improving capabilities of DeepSeek and limited commercialization strategies are redefining industry benchmarks as "high performance, low cost," thereby squeezing the profit space of independent model providers. Commercialization of AI is still dominated by enterprises, and cloud and Model as a Service (MaaS) are the clearest paths. In the consumer AI field, not all users can profit equally. The shift from "prosumers" with high intent value to mass commercialization achieved through subscriptions is still in its early stages. Models based on advertising/trading may offer longer-term revenue options. Bank of America expects that as investor focus shifts to computing power and ecosystem strength, the overall performance advantage of AI labs compared to internet giants will gradually weaken. Therefore, a series of events will drive stock performance in the second half of 2026. Between the two, the premium of KNOWLEDGE ATLAS reflects its faster annual recurring revenue (ARR) growth, stronger talent density/public support, and its leading position in enterprise revenue. However, its expected market-to-sales ratio for 2027 is as high as 52 times, while MINIMAX is only 19 times, which seems too high relative to product breadth. The bank believes that after the end of the lock-up period (July 8th), MINIMAX is expected to see catch-up trades, benefiting from its potential inclusion in the "Southbound Channel" in early August and the catch-up of corporate revenue after the release of M3. Bank of America believes that Alibaba Group Holding Limited Sponsored ADR remains the best representative of the "China AI/cloud computing" theme. The bank is also optimistic about TENCENT (00700) accelerating its development in the model/consumer agent space. Given ByteDance's and telecommunications companies' active capital expenditures and their growing influence in the AI model/domestic chip ecosystem, they continue to be key drivers of volatility.