云天励飞董事长陈宁：聚焦AI芯片战略

Core Viewpoint - The company, CloudWalk Technology, is strategically focusing on AI chips, aiming to build a domestic computing "accelerator" centered around edge computing, cloud-based large model inference, and embodied intelligence [1][2]. Company Development - Founded in 2014, CloudWalk Technology has spent 11 years developing AI applications across consumer, enterprise, and industry sectors, achieving multiple innovations [1]. - The company launched its second-generation neural network processor (NPU) in 2018, which has been widely applied in various fields such as smart cameras and industrial intelligence [1][2]. - In 2023, the company released its third-generation chip, DeepEdge10, designed for large model inference, featuring a modular architecture that supports scalable computing systems [2][3]. Market Outlook - The chairman believes that 2025 will mark a significant turning point for AI, with a shift from training to inference, leading to an explosive growth in inference computing demand [2][3]. - Although the current AI inference chip market is relatively small compared to the training chip market, it is expected to grow at a much faster rate in the next 3 to 5 years [3]. Key Application Scenarios - The company is focusing on three main application scenarios: edge computing, cloud-based inference acceleration, and embodied intelligence, aiming to create a cost-effective AI inference chip technology and product system [4][5]. - In edge computing, the company has implemented several demonstration projects and gained recognition from leading industry clients, leveraging its expertise in low power consumption and high performance [4]. - For cloud inference acceleration, the company has developed a reasoning acceleration card based on domestic technology, enhancing flexibility and efficiency in chip design and deployment [4][5]. - In the field of embodied intelligence, the company has partnered with over ten robotics manufacturers, with its edge chip products already in practical deployment [5].