en.Wedoany.com Reported - On June 16, the Artificial Intelligence and Ecology Conference (AIEC 2026) opened at the Zhongguancun Exhibition Center in Beijing. Hosted by the Global Industry Research Institute of Tsinghua University, the conference focused on topics such as the development of intelligent agents, computing power infrastructure construction, and industrial ecosystem co-building.
TECHPRO, as a domestic AI computing power enterprise, was invited to attend. The company's Chief Architect, Xia Zhongmou, delivered a speech titled "TECHPRO's Domestic AI Computing Power Token Service Practices," sharing practical experiences and industry insights on the large-scale deployment of domestic computing power and the commercial implementation of Token services.
Xia Zhongmou pointed out that the deployment of domestic computing power currently faces three core challenges: First, the service capability of large-scale clusters needs improvement. Domestic computing power still requires breakthroughs in single-card performance, interconnect bandwidth, software ecosystem, and large-scale operations and maintenance for pre-training clusters at the thousand-card and ten-thousand-card levels. Second, there is the issue of adapting Agentic AI computing power architectures. In interactive computations such as planning and scheduling, the CPU proportion significantly increases, posing challenges to traditional CPU/GPU ratio architectures. Third, there are barriers to ecosystem migration. Mainstream frameworks and operator libraries are deeply tied to the CUDA ecosystem, making the migration of models to domestic computing power complex, costly, and time-consuming.
To address these pain points, TECHPRO leverages its high-performance computing expertise to offer comprehensive solutions.
In terms of cluster capabilities, TECHPRO continues its experience in large-scale parallel computing technology to ensure stable operation and efficient acceleration ratios of supercomputing clusters. The company's team has won the "Gordon Bell Prize," the highest international award in high-performance computing, three times, and has deep expertise in large-scale parallel computing at the million-core level. Currently, its thousand-card cluster capabilities have matched international levels, and breakthroughs in ten-thousand-card and hundred-thousand-card cluster technologies are accelerating.
At the architecture level, TECHPRO has developed a self-designed on-chip integrated heterogeneous many-core architecture. A single chip integrates general-purpose computing, data processing, and accelerated computing cores, enabling on-chip collaborative computing. This significantly reduces I/O overhead, greatly improves CPU and GPU collaboration efficiency and hardware utilization, and meets the complex computing power demands of intelligent agents.
In the ecosystem domain, the company offers diverse development paradigms for different developers, providing rich tool components to enable rapid migration of code from mainstream frameworks such as PyTorch and vLLM. Currently, it can meet 80%-90% of mainstream scenario adaptation needs.
At present, TECHPRO has implemented intelligent computing center projects in multiple locations across China. Computing power nodes in Zhengzhou, Yancheng, Yan'an, and Lihu Future City have been built and put into operation, with a total computing power scale reaching thousands of PFlops. Among them, the Henan Airport Intelligent Computing Center has officially begun providing Token services with over 1300P of intelligent computing power. The company adopts a collaborative model involving "computing power vendors, ISVs, and end customers," creating a number of benchmark cases: building an intelligent social security audit system for local health commissions, leveraging the Qwen3-32B model to achieve efficient medical insurance data auditing and risk screening; collaborating with Tsinghua University to create a "digital employee" for code generation, improving software development efficiency by 30%-50%; and implementing projects in data synthesis, intelligent operations and maintenance inspections, intelligent testing, and AI training platforms, covering fields such as government services, scientific research, enterprise services, and education and training.
Xia Zhongmou stated that over 80% of domestic application scenarios have low sensitivity to latency. With its high cost-effectiveness and autonomous controllability advantages, domestic computing power can fully meet mainstream Token service demands. In the future, TECHPRO will continue to refine computing power technology, improve the industrial ecosystem, deepen cooperation with industry chain partners, and persistently promote the large-scale deployment of domestic AI computing power to support the high-quality development of China's artificial intelligence industry.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
