First AI product: “Jinghong Series” launched. On March 12, Jingjiawei announced that the company has successfully developed the Jinghong series of high-performance intelligent computing modules and complete machine products for application fields such as AI training, AI reasoning, and scientific computing, and will be promoted to the market as soon as possible. We believe that as a leading domestic GPU company, Jingjiawei has enriched its product line by releasing its first computing power product for the AI field. While increasing the company's overall competitiveness, the AI series products are expected to become a new driving force for the company's performance growth in the future.
“Jing Hong” + “JM”, AI and graphics fight with both fists. Jingjiawei's JM series products mainly involve graphics processing. They have experienced iterations of the 5/7/9 series, and each version has segmented products suitable for different scenarios. Among them, the JM9-1 and JM9-2 products both have 8GB of video memory capacity, and the video memory bandwidth is 128GB/S and 25.6Gb/s, respectively. According to the company's announcement, the Jinghong series includes high-performance intelligent computing modules and complete machine products. We believe that in the future, Jinghong series products may continue the tradition of the JM series, maintain iterative versions while presenting multiple segments of products for different application scenarios.
It supports mainstream ecosystems at home and abroad, and has strong compatibility. According to the company announcement, Jinghong series products support mixed precision computation such as INT8, FP16, FP32, FP64, etc., support new multi-card interconnection technology to expand computing power, and are also compatible with mainstream domestic and foreign CPU, operating system and server vendors. They can support current mainstream computational ecosystems, deep learning frameworks and algorithm model libraries, and greatly shorten the user adaptation verification cycle.
We believe that since the current mainstream AI computing ecosystem and system architecture are relatively mature, and while it is difficult to rebuild from the underlying architecture, the compatibility between domestic computing power hardware and mainstream ecosystems and systems is particularly important. According to the company announcement, Jinghong series products will be able to adapt and be compatible with mainstream systems and ecosystems at home and abroad, and may become one of the core competencies of the Jinghong series in the future.
Multi-card interconnection expands computing power and breaks single-card performance bottlenecks. According to the company's announcement, Jinghong series products will support new multi-card interconnected computing, thereby expanding computing power. Currently, the computational demand for AI and high-efficiency computation is increasing, and under the training requirements of large-scale LLM models, there are also higher requirements for GPU performance and connection speed. We believe that the company's multi-card interconnection technology will break the bottleneck of single card performance to a certain extent, and that the connected AI computing power series products will meet the needs of more complex scenarios.
Profit forecasting and investment advice. We expect Jingjiawei's 2023E/2024E/2025E to achieve operating income of 7.5/12.5/1.66 billion yuan and net profit to mother of 0.6/2.9/40 billion yuan, respectively. The PS valuation corresponding to the current stock price is 47.5/28.5/21.5x, respectively, maintaining the “Highly Recommended” rating.
Risk warning: Order fulfillment progress falls short of expectations, new product development falls short of expectations, downstream demand falls short of expectations.