JPMorgan released a research report stating that the launch of DeepSeek V4 has removed the most significant competitive negative catalyst from April to May. The catalyst path for the next two months has shifted to asymmetrically positive, including index inclusion (review in May, effective in June), KNOWLEDGE ATLAS's inclusion in the Southbound Stock Connect (early June), and the releases of GLM-5.5 and MiniMax M3. The firm recommends accumulating shares on pullbacks, maintaining an "Overweight" rating on KNOWLEDGE ATLAS (02513) and MiniMax-W (00100), with target prices of HK$950 and HK$1,100 respectively. Following the release of the DeepSeek V4 preview last Friday (24th), the stock prices fell by 9%, which the market interpreted as a competitive impact on listed large language model companies. JPMorgan believes this is an overreaction; conversely, the V4 launch represents a tailwind for the industry rather than a zero-sum competitive shock. The release strengthens three pillars of the firm's framework: unlocked computing power supply, pricing discipline, and structural cost curve compression, while relative competitive positioning remains intense rather than rebalancing. The report indicated that DeepSeek V4's use of Huawei's Ascend chips demonstrates the practical feasibility of domestic chips for cutting-edge AI model inference. As domestic chips become more widespread, KNOWLEDGE ATLAS and MiniMax will gain more cost-effective solutions, enhancing their ability to convert token demand into billable revenue. V4's Pro/Flash tiered pricing (with an output price gap of approximately 12x) also validates the task-based monetization logic, which is consistent with the existing strategies of KNOWLEDGE ATLAS and MiniMax.
Comments