UBS Securities: The three major themes of the Chinese AI model worth paying attention to in the second half of the year are model capability, monetization, and ROI

The Zhitong Finance App learned that Xiong Wei, a Chinese internet industry analyst at UBS Securities, pointed out that after the launch of Zhi Spectrum (02513) and MiniMax (00100) at the beginning of this year, investors' interest in Chinese AI models has increased, and discussions on model iteration, monetization, and competition have also deepened. Currently, the three major themes of the Chinese AI model worth paying attention to in the second half of the year are model capability, monetization, and ROI.

He pointed out that although it is expected that model intelligence, which is dominated by programming ability, will continue to improve, he believes there are early signs that the top models may be able to defend their leading position, mainly due to data flywheels, increased user stickiness brought about by better commercialization, and an increase in AI adoption rate in model development (such as recursive self-improvement).

The bank expects model layer monetization to continue to grow as the potential market space for programming expands from developers to a wider range of groups, and progress in multi-modality aspects may be underestimated. Furthermore, the bank also anticipates that the term ROI will receive more attention. Enterprises and users will shift from maximizing word usage to word optimization, which will drive hierarchical demand and expand the differences in pricing capabilities between the top model and other models. Since the Chinese model is structurally cost-effective, it is expected that it can benefit from this.

UBS Securities reaffirms that programming is the core ability that supports the level of model intelligence and has a clear path to expand the intelligent boundaries of the model. It believes that AI programming is one of the most suitable post-training fields based on reinforcement learning. It has noticed that emerging Chinese AI laboratories are more active in constructing programming intelligence products/frameworks to promote commercialization, which is expected to form a potential data flywheel. It is expected that models with the highest programming capabilities can further strengthen their position.

In terms of monetization, Xiong Wei anticipates that the potential market space for AI programming will continue to expand. As enterprise applications shift from maximizing word usage to ROI constraints, he is more optimistic about the potential for increasing the global share of the Chinese model. However, he also reminded that the supply of computing power is still the main bottleneck in ARR's growth rate.