CMSI: China Large Model Industry Agent Commercialization Unstoppable; Initiates KNOWLEDGE ATLAS and MINIMAX-W at Overweight

CMSI released a report stating that the commercialization flywheel of Agents in Chinas large model industry is unstoppable. It initiated coverage on KNOWLEDGE ATLAS (02513.HK) -39.500 (-4.394%) Short selling $10.06M; Ratio 1.365% and MINIMAX-W (00100.HK) -30.000 (-4.071%) Short selling $49.43M; Ratio 7.366% with Overweight ratings. The two companies possess distinct market positioning and competitive moats and are currently in an accelerated commercialization phase, while the large model industry is entering an explosive stage of Agent capabilities.The report noted that over the past few years, Chinas large model industry has experienced rapid development, evolving from conversational AI to multimodal integration, and from enhanced reasoning to improved Agent capabilities, with continuous advancements in model intelligence and technical paradigms. Chinas average daily token consumption surged from 100 billion in early 2024 to 140 trillion in March, representing more than a thousandfold increase within two years, primarily driven by the exponential growth in calls brought about by continuous Agent deployment. The global large model market size is projected to grow from USD10.7 billion in 2024 to USD206.5 billion in 2029, representing a CAGR of 80.7%. Chinas AI market is expected to reach RMB993 billion by 2030, with a CAGR of 35.5% from 2024 to 2030, while the enterprise-level large model market is expected to record a CAGR of 64%.The report indicated that global competition in large models has narrowed to a China-US rivalry. According to Frost & Sullivan, China and the US account for 86% of the global total number of large models. The US maintains leadership in foundational algorithms and chips, but the gap between Chinas advanced models and those of the US is rapidly narrowing. The US leads in closed-source models, while China leads in open-source weight models. Competition in Chinas large model industry is no longer limited to benchmark performance. Model intelligence determines vendors upper growth ceiling and brand influence; multimodal capabilities open new battlegrounds in content creation and industry applications; and architectural efficiency directly affects commercialization costs, corporate profitability, and application accessibility. In the future, vendors that can strike a balance across these three dimensions and deeply integrate them with their ecosystems and business models are more likely to prevail in an increasingly differentiated competitive landscape.The report stated that KNOWLEDGE ATLAS and MINIMAX-W, as representative pure-play large model companies, both focus on full-modality AI deployment. Leveraging advantages such as high technological focus, short decision-making chains, and strong neutrality, they form differentiated competition against comprehensive players such as Alibaba and ByteDance.The broker remains positive on KNOWLEDGE ATLAS, which focuses on pursuing the upper bound of model intelligence, continuously strengthening technological barriers and commercialization loop closure in programming capabilities, and enhancing model pricing power. Revenue CAGR for 2025-2028 is forecast at 172%, with adjusted net loss margin expected to narrow significantly from -439% in 2025 to -14% in 2028. Based on SOTP valuation, the TP is HKD1,282, corresponding to 68x/35x FY27/FY28 forward PS.The broker also remains positive on MINIMAX-Ws market positioning centered on full-modality in-house development and extreme cost efficiency, with focus on the upcoming major iteration of the M3 model. Revenue CAGR for 2025-2028 is projected at 193%, with adjusted net loss margin expected to narrow sharply from -317% in 2025 to -12% in 2028. Based on SOTP valuation, the TP is HKD982, corresponding to 43x/20x FY27/FY28 forward PS. (ha/u)(HK stocks quote is delayed for at least 15 mins.Short Selling Data as at 2026-05-11 16:25.)