In recent years, China's artificial intelligence has accelerated its industrial application, and the construction of computing infrastructure has continued to advance. With the continuous development of large model training, inference, and intelligent agent applications, data centers have put forward higher requirements for computing power, networking, storage, and scheduling capabilities.
During this process, the data transmission efficiency within the AI computing power cluster is becoming a new challenge, and the importance of DPU (Data Processor) as a key chip connecting computing power, network, and storage is becoming increasingly prominent. Faced with this pattern, a group of DPU enterprises with strong research and development capabilities and broad application prospects have emerged in China, accelerating their leap from technological breakthroughs to large-scale applications, becoming an important force in filling the gaps in the computing power base.
AI Reshaps Global Computing Landscape, DPU Becomes New Focus of Computing Infrastructure
Since OpenAI launched ChatGPT, the global artificial intelligence industry has entered a new phase of rapid development. The training and inference of AI large models not only rely on the intensive computing power of GPU and other computing chips, but also pose higher requirements for data transmission, resource scheduling, and system collaboration within the computing power cluster.
It is widely believed in the industry that the competition in the AI era appears to be a competition of models and applications, but at the bottom it is actually a competition of computing power infrastructure. Without stable, efficient, and autonomously controllable underlying chips and infrastructure, model training, inference services, industrial applications, and data security will all be constrained.
Currently, the computing power system of data centers is supported by three major chips: CPU (central processing unit), GPU (graphics processing unit), and DPU. Among them, the CPU undertakes general computing tasks, the GPU undertakes parallel computing and AI acceleration tasks, and the DPU plays the role of the "invisible commander" within the computing power cluster. In the context of general computing, DPU can offload tasks such as network, storage, security, and virtualization from the CPU side, freeing up valuable general-purpose computing resources; In intelligent computing scenarios, DPU can provide GPU clusters with high bandwidth and low latency data transmission capabilities, alleviate network congestion, reduce GPU waiting time for data, and significantly improve the effective computing power utilization of AI clusters.
Based on this value proposition, DPU, as a key link connecting computing power, networks, and data flow, is becoming a new focus in the construction of computing infrastructure. Among the six core chips of Nvidia's new generation AI intelligent computing system Vera Rubin, four of them focus on solving network bottlenecks in computing power clusters, two of which are DPU products, demonstrating its core position in the AI computing power system.
At the policy level, in October 2023, the Ministry of Industry and Information Technology and six other departments issued the "Action Plan for High Quality Development of Computing Infrastructure", which clearly stated the need to "optimize the efficient transportation quality of computing power", carry out technological upgrades and pilot applications such as DPU and lossless networks, and achieve high-performance transmission of computing power center networks. Currently, with the acceleration of the construction of AI intelligent computing centers and cloud data centers, this niche field is accelerating in industrial practice.
From the perspective of long-term industry development, core chips such as CPUs, GPUs, and DPUs are related to the autonomous and controllable computing infrastructure. Among them, the CPU track in China has formed mature independent product lines such as Haiguang, Huawei Kunpeng, and Feiteng, and achieved large-scale implementation; On the GPU track, Cambricon, Huawei Ascend, Alibaba Pingtou Brother, as well as a group of domestic manufacturers such as Moore Thread, Muxi, Boren, and Haiguang, continue to iterate their products to accelerate the replacement process. In contrast, DPU, as an important part of data center infrastructure, has long been dominated by overseas enterprises, and the domestic ecosystem is still in the stage of accelerating replenishment. In recent years, domestic DPU enterprises such as Yunbao Intelligence have begun to focus on the research and development of high-performance products and large-scale applications, which is changing the industry landscape.
Focusing on the entire process, breaking through real-life scenarios, and testing product capabilities
It is reported that in the wave of accelerated development of domestic DPUs, Yunbao Intelligence is accelerating the breakthrough of key technologies and industrialization landing. According to public information, Yunbao Intelligence was established in August 2020 and is the first DPU enterprise in China to achieve a breakthrough in 400Gbps speed technology. The first generation DPU chip was successfully taped out in one go, and this large chip composed of 20 billion transistors can be directly commercialized on a large scale without modifying any transistors (i.e. A0 version). In contrast, overseas top manufacturers of DPU with the same specifications have gone through three generations of product iterations and took 7 years to achieve commercial implementation, while Yunbao Intelligent completed the full process breakthrough from design, verification to mass production in two and a half years. It is reported that the Yunbao Intelligent DPU series products have entered the "Made in China 14th Five Year Plan Achievement Exhibition" at the National Museum of China, and have been included in the "National Treasures" display array. They have become one of the three chip companies exhibited in the exhibition and the only DPU company selected.
DPU belongs to system level chips, and whether enterprises can truly enter real data center scenarios is the core criterion for testing product capabilities. At the same time, DPU's customer verification not only depends on chip parameter indicators, but also on software stack completeness, ecological compatibility, system stability, supply guarantee capability, and long-term operational performance in real business scenarios. It is reported that Yunbao DPU products have achieved large-scale deployment in cloud computing, Internet, finance, operators, power grids and other scenarios, and have delivered more than 100000 pieces in total.
From the perspective of industry development laws, it is a common phenomenon for hard technology enterprises to have a high concentration of customers in the early stages of growth. System level chips like DPU, which were able to undertake large-scale validation and deployment in the early days, were often concentrated in large cloud vendors, operators, and top industry customers - products need to be validated in complex scenarios first, and then gradually replicated and promoted to more customers and application scenarios. It is understood that Yunbao is promoting the diversification of customer structure, and has established cooperation with operators, Internet enterprises, AI large model companies, and customers in finance, power and other industries, and its application scenarios continue to expand.
Focus on long-term value, domestic DPUs embark on the path of independent innovation
The high-end chip industry generally has the characteristics of high R&D investment, long industrialization cycle, and lagging commercialization. For large chip companies such as DPU, GPU, CPU, etc., it is usually necessary to first complete core technology breakthroughs and product validation, then enter the stage of large-scale commercial implementation, and finally gradually release profitability.
An industry insider analyzed that judging the commercialization quality of hard tech chip companies should not only focus on short-term profit indicators, but also pay attention to whether the technology has entered the real scene, whether the revenue comes from core products, whether customers have completed scale verification, and whether the product has the ability to continue iteration.
From financial data, Yunbao Intelligent's operating revenue increased from 173200 yuan in 2023 to 369891700 yuan in 2025, with a year-on-year growth of over 900%. DPU product revenue has become the main source of main business revenue, indicating that the core product has entered the stage of large-scale commercialization. At the same time, the company is still in a period of high investment in research and development. In 2025, the company's net profit attributable to the parent company will incur a loss of approximately 1.189 billion yuan, including approximately 643 million yuan of one-time recognized share based payment expenses (mainly from accounting provisions for equity incentives). After excluding non recurring gains and losses, the net loss attributable to shareholders in 2025 is approximately RMB 562 million. The funds are mainly used for chip research and development, and the scale of the loss has narrowed compared to 2024.
In terms of market space, Sullivan's report predicts that the global DPU market size will grow from 196.491 billion yuan in 2025 to 436.239 billion yuan in 2030, and the Chinese DPU market size will grow from 49.758 billion yuan in 2025 to 129.91 billion yuan in 2030. The domestic DPU market has a wide space for domestic substitution.
Industry analysis believes that sustained high R&D investment and early stage losses are the main characteristics of the early development of large chip enterprises. For Yunbao Intelligence, short-term large R&D investment brings stage financial pressure, but long-term industry value and market prospects are worth paying attention to.
Currently, AI infrastructure is shifting from single machine computing power to cluster computing power, and from single point chip capability to system efficiency, which constitutes the core driving force for the growth of the DPU market. The long-term demand for AI infrastructure and data center construction in China, coupled with the industry trend of high-performance core chips being independently controllable, has also opened up a development window for domestic DPU enterprises. It is reported that Yunbao has launched 400Gbps high-performance full function DPU and AI DPU, and continues to invest in the next generation 800Gbps/1.6Tbps products. On the enterprise side, it is hoped that through independently developed DPU products and technology platforms, intelligent computing centers, cloud data centers, and industry data centers can comprehensively improve system efficiency and lay a solid foundation for the domestic computing power base.