实控人提议回购公司股份，将用于员工持股计划与股权激励本次回购股份的原因和目的：陈天石博士基于对公司未来发展的信心和对公司价值的认可，为了维护广大投资者利益，增强投资者对公司的投资信心，同时促进公司稳定健康发展，有效地将股东利益、公司利益和员工个人利益紧密结合在一起，提议公司以自有资金回购部分股份，并将回购股份用于员工持股计划或股权激励。Incidents:
On July 18, 2024, the company announced that Dr. Chen Tianshi, the controlling shareholder, actual controller, chairman and general manager of the company, proposed that the company use its own funds to repurchase some of the company's issued A shares through centralized bidding transactions. The total repurchase capital is 20 million yuan to 40 million yuan. The repurchase period is 12 months from the date the company's board of directors reviewed and approved the share repurchase plan.
Investment highlights:
The actual controller's proposal to repurchase the company's shares will be used for employee stock ownership plans and equity incentives: Dr. Chen Tianshi is based on confidence in the company's future development and recognition of the company's value. In order to maintain the interests of investors, enhance investors' investment confidence in the company, and promote the stable and healthy development of the company, effectively closely integrate shareholders' interests, company interests, and employees' personal interests, he proposed that the company use its own funds to repurchase some of the shares, and use the repurchase shares for employee stock ownership plans or share incentives.
Continuing to focus on the development of general-purpose AI chips, the company continues to promote iterative optimization of intelligent processor microarchitectures and instruction sets, focusing on optimizing training and inference scenarios. The next-generation intelligent processor microarchitecture and instruction set will focus on optimizing scenarios such as large models for natural language processing, large models for video image generation, and training and reasoning for large models of recommended systems. At the same time, the company continues to promote the development of training and inference software platforms:
(1) Training software platform: The company continues to increase the number of support for frame operators, improve the development of training software stacks for large models, and carry out fine-tuning and pre-training support work for multiple large models. It has achieved strong competitiveness in terms of the performance and accuracy of large model training, and has reached the leading level of computational efficiency in the industry.
(2) Inference software platform: The company has developed basic software to support large language models and multi-modal AIGC inference services, and has continuously optimized the performance of traditional models and generative large models, which has accelerated actual application deployment. The company has developed BangTransformer, a distributed inference acceleration library, and large model inference has achieved industry-leading model calculation efficiency.
The construction of software ecosystem platforms continues to be strengthened, and domestic AI computing power demand is increasing at an accelerated pace. On July 3, 2024, the China Academy of Information and Communications Technology and others officially launched the “Intelligent Computing Chip Application Ecosystem Construction” project. More than 40 companies from the fields of intelligent computing chips, complete machines, and applications conducted exchanges and discussions on the construction of an intelligent computing chip application ecosystem. Representatives of companies such as Zhongke Cambrian shared on topics such as intelligent computing chips and large model applications, domestic intelligent computing chip implementation challenges, and full-function GPUs to accelerate the application of artificial intelligence industries.
Demand for domestic AI computing power is increasing at an accelerated pace. According to the Financial Services Association on June 27, Internet vendors have tried domestic AI chips, and customers with high security requirements, such as finance, government, and enterprises, all use domestic chips. In July, China Telecom issued a server procurement announcement. Among them, the number of G series (domestic series) reached 0.1053 million units, accounting for 67.5%, and the localization ratio reached a record high.
Profit forecast and investment rating: The company is a leading enterprise in the field of smart chips in China. It can provide serialized smart chip products and platform-based system software with cloud-edge integration, software and hardware collaboration, training and reasoning integration, and a unified ecosystem. It is expected to benefit from the computing power requirements brought about by the development of large models and the continuous iteration of the company's products. We forecast revenue of 1.4/2.493/3.484 billion yuan for 2024-2026, corresponding PS of 75.59/42.45/30.37X, and net profit to mother of -0.441/0.018/0.25 billion yuan, respectively, maintaining the “gain” rating.
Risk warning: The macroeconomy affects downstream demand, the development of the large model industry falls short of expectations, market competition intensifies, the game between China and the US intensifies, the company's performance falls short of expectations, the company's new product development falls short of expectations, etc.
