A-share subscription | Suiyuan Technology (688801.SH), one of the “Four Dragons of Domestic GPUs”, opens subscription today

The Zhitong Finance App learned that on September 2, Suiyuan Technology (688801.SH) began a subscription. The issue price was 142.18 yuan/share, and the subscription limit was 0.65 million shares. It belongs to the Shanghai Stock Exchange, and CITIC Securities is its sponsor. Suiyuan Technology, Moore Thread, Mu Xi Co., Ltd., and Bizao Technology are collectively known as the “Four Dragons of Domestic GPUs.” Previously, Moore Thread, Mu Xi Co., Ltd., and Bizao Technology had closing gains of 425.46%, 692.95%, and 75.82% on the first day of listing, respectively.

According to the prospectus, the company is one of the leading enterprises in the field of cloud AI chips in China and is committed to becoming a “leading enterprise in general artificial intelligence infrastructure”. The company adheres to the original innovation and independent research and development technology route to build a core competitiveness and moat for long-term sustainable development. Since its establishment 8 years ago, the company has developed and iterated 5 cloud-based AI chips with four-generation architectures, and built a complete product system covering AI chips, AI acceleration cards and modules, intelligent computing systems and clusters, and AI computing and programming software platforms.

In the core technology field, after years of accumulation, the company has formed a comprehensive and three-dimensional core technology system with three categories: chips and hardware, software and programming platforms, and computing power cluster solutions. In terms of underlying hardware, the company is based on an autonomous instruction set, benchmarking Nvidia's TensorCore accelerated computing unit and NVLink card interconnection technology, the original autonomous GCU-CARE accelerated computing unit and GCU-LARE high-speed interconnect technology. The corresponding architecture not only has programming flexibility, but also deeply supports highly parallel accelerated computation of large AI models. At the software platform level, the company did not follow the CUDA ecosystem led by Nvidia and developed its own “TopsRider”, a full-stack AI computing and programming software platform including drivers, compilation languages and compilers, operator libraries, and tool chains to link the company's hardware and artificial intelligence applications, greatly reducing the difficulty and migration costs of programming development and migration costs based on mainstream AI models based on the company's hardware, so that the company's hardware products can better release performance in application scenarios. In terms of computing power clusters, the company's kilocalories and 10,000 cards intelligent computing center projects have achieved revenue during the reporting period. At present, the company has collaborated with customers to develop supernode solutions and jointly build a 10,000 card high-speed internet cluster with commercial value.

In terms of ecological construction, it has benefited from years of deep cooperation with major Internet companies in software and hardware customization. The company's multi-generation products have been commercialized on a large scale in a wide range of Internet AI scenarios, and continues to provide AI computing power support for national-level Internet applications based on traditional AI models to AI models. The company's continuous iterative product development capabilities and product competitiveness have been tested and recognized by the market, and is gradually achieving a key leap from “closed loop of technical products” to “closed loop of commercial value”. Furthermore, the company is also making full use of its strengths to build an ecosystem with a wider range of downstream partners. In addition to participating in the intelligent computing center project of the national “East Digital and Western Computing” hub node, the company is actively deepening cooperation with domestic telecom operators and developing business opportunities in various vertical industries to provide inclusive computing power support for AI to empower thousands of industries. Furthermore, in order to achieve sustainable development, Suiyuan Technology attaches great importance to long-term strategic cooperation with partners in the entire industry chain, including EDA/IP, wafer manufacturing, packaging testing, and system components, to ensure stable product development and supply delivery.

The domestic cloud AI chip industry is in the early stages of development. The international manufacturer Nvidia occupies a major share of the Chinese market, but local Chinese companies continue to break through technical barriers and occupy a certain market share. Local cloud AI chip manufacturers in China include DSA architecture manufacturers represented by Huawei HiSilicon, Cambrian, and companies, and GPGPU architecture manufacturers represented by Moore Thread, Mu Xi Co., Ltd., Tianshu Zhixin, and Bizao Technology.

According to IDC data and the company's sales volume, the overall shipment scale of AI acceleration cards in China in 2025 is about 4 million, of which Nvidia accounts for about 55% of the market share with about 2.2 million shipments. The company sold 66,000 AI accelerator cards and modules in the same year, corresponding to a 1.7% share of the AI accelerator card market in China, ranking among other domestic AI chip manufacturers. In the future, there is some uncertainty about Nvidia's sales in the domestic market. It is a probable trend that domestic AI chip manufacturers continue to increase their share of shipments in the Chinese market.

On the financial side, in 2023, 2024, and 2025, the company achieved operating income of approximately RMB 301 million, RMB 722 million, and RMB 990 million respectively; during the same period, net losses were approximately RMB 1,665 million, RMB 1,510 million, and RMB 1,164 billion, respectively.