According to the report of China Media Group, Cambrian issued an announcement on the 17th, announcing the adjustment of the 2025 A-share stock plan to specific targets. According to the latest plan, the company's stock issuance amount shall not exceed 20.9175 million shares, and the total amount of funds raised shall not exceed 3.9885.3273 million yuan (RMB, the same below). Cambrian said that after deducting issuance fees, the funds raised will be mainly used for chip platform projects for large models, software platform projects for large models, and replenishing working funds.
Cambrian pointed out that this fundraising investment project is aimed at the development needs of large-scale model technology and application. Based on the hardware architecture characteristics of the company's smart chips, it will carry out relevant innovative research on optimization strategies, software algorithms and software tools in the areas of key demands for large-scale models such as high degree of mobility, high computing efficiency, and high storage efficiency; and build large-scale models-oriented innovative research. Software platform, the platform will cover three functional modules: flexible editing system, training platform and reasoning platform, to improve the ease of use and adaptability of the company's smart chips, and support the development of services from large-scale algorithms to the full business process of application deployment. Through the construction of this project, the company can also open an easy-to-use software platform for a broad AI industry.
Cambrian continues to point out that the company has successively developed a series of excellent smart processors and smart chip products, including Cambrian 1A, Cambrian 1H, and Cambrian 1M series smart processors for the final scene; cloud smart acceleration card series products based on Siyuan 100, Siyuan 270, Siyuan 290, and Siyuan 370 chips; and edge smart acceleration card based on Siyuan 220 chips.
In the successful commercial use of these products, the company has accumulated rich technical advantages and business experience. The company's smart chips and processor products can efficiently support large-scale model training and reasoning, vision, voice processing and natural language processing, as well as multi-mode AI tasks that integrate technologies such as recommendation systems, and can support the training and reasoning tasks of mainstream sourced large-scale models in the market, including the DeepSeek series, LLaMA series, GPT series, BLOOM series, GLM series and multi-mode (StableDiffusion), etc.