清微智能获超20亿元融资 已启动上市筹备工作

Core Insights - Qingwei Intelligent, an AI chip company based in Beijing, has completed a C-round financing of over 2 billion RMB and is preparing for an IPO, aiming to establish itself as a benchmark in the non-GPU reconfigurable computing architecture sector [1][3] Financing and Investment - The financing round was led by Beijing state-owned enterprise Jingneng Group, with participation from multiple institutions including Beichuang Investment, Jiantou Investment, Wuyuefeng Science and Technology Innovation, and Chengdu Science and Technology Investment [3] - The funds will primarily be used for the R&D of next-generation reconfigurable chip core technologies, large-scale deployment of intelligent computing scenarios, and the recruitment and training of high-end technical talent [3] Technology and Product Development - Qingwei Intelligent has developed the TX81 reconfigurable AI chip, which utilizes "C2C computing grid technology" to achieve high bandwidth and low latency data flow [3] - The REX1032 training and inference integrated server, powered by this chip, can reach a computing power of 4 PFLOPS and supports trillion-parameter large model deployments, offering approximately 50% cost reduction and 3 times efficiency improvement compared to traditional solutions [3] Market Position and Commercialization - The company's products are compatible with major large models such as DeepSeek, Qwen, and SenseTime's Riri New, and it has established large-scale intelligent computing clusters in various regions including Zhejiang, Beijing, and Anhui [3] - Qingwei Intelligent has received over 20,000 orders for reconfigurable computing cards and is projected to rank sixth in domestic AI acceleration card shipments by mid-2025, placing it in the top tier of domestic computing acceleration cards [3] Leadership and Recognition - The founder and CEO of Qingwei Intelligent, Wang Bo, has recently been awarded the title of "2025 China IC Design Industry Annual Entrepreneur," recognized for his forward-looking layout and continuous advancement in the localization of new architecture AI chips [4]