Cambricon(688256) 民生证券·2024-09-02 14:00Investment Rating - The report maintains a "Recommend" rating for Cambricon-U (688256 SH) [2][3] Core Views - Cambricon-U is a leading domestic AI chip company, actively developing new-generation intelligent processor microarchitectures and instruction sets, and continuously optimizing and iterating its inference and training software platforms [2] - The company's chip products have established cooperation intentions with leading companies in multiple industries, and its brand effect is gradually becoming prominent [1] - The company's training software platform has been able to fully support the distributed training needs of mainstream large models, and the inference software platform has achieved coverage of mainstream AIGC business scenarios such as text-to-text, text-to-image, text-to-video, and multimodal [2] Financial Performance and Forecast - In the first half of 2024, the company's operating income was 647 653 million yuan, a year-on-year decrease of 43 42%, with a net loss attributable to the parent company of 530 million yuan and a net loss attributable to the parent company after deducting non-recurring gains and losses of 609 million yuan [1] - The company's R&D investment in the first half of 2024 was 4474 76 million yuan, accounting for 690 92% of operating income [1] - The company's operating income is expected to be 1 536 billion yuan, 3 544 billion yuan, and 5 015 billion yuan in 2024, 2025, and 2026, respectively, with corresponding PS ratios of 69X, 30X, and 21X [2][3] - The company's net profit attributable to the parent company is expected to be -257 million yuan, 147 million yuan, and 322 million yuan in 2024, 2025, and 2026, respectively [3][6] Software Platform and Ecosystem - The company's training software platform has updated its support for PyTorch 2 1/2 3, supported multiple training and inference business scenarios, and achieved a long-term mechanism for quickly following community versions [2] - The training software platform has added support for mainstream networks such as Llama3 series and Qwen series, and has optimized parallel training functions [2] - The inference software platform has supported and optimized a series of text-to-text models such as Llama series and Qwen series, as well as mainstream multimodal models such as StableDiffusion3, achieving coverage of mainstream AIGC business scenarios [2] - The company has developed a PyTorch third-party operator acceleration library, achieving leading operator calculation efficiency in the industry [2] Industry and Market Position - The AI chip industry is highly competitive, and technological leadership is the core foundation for the company to occupy the market and achieve long-term development [1] - The company focuses on the development of network information business, and deeply promotes the continuous development of independent products, industry solutions, and service-oriented business with network information business as the core [1]