The Next Phase of Computing Infrastructure: From National Network to Massive Clusters
The recently released "15th Five-Year Plan" for the information and communications technology sector by the Ministry of Industry and Information Technology outlines a comprehensive strategy for advancing China's computing power infrastructure. The focus extends beyond mere capacity expansion, emphasizing a systematic and sovereign approach to building next-generation computational resources.
A Hierarchical Computing Architecture
A cornerstone of the plan is the acceleration of a national integrated computing power network. This network is designed as a multi-tiered system, structured around "hub-region-edge" layers. This architecture aims to intelligently distribute computing tasks based on latency, cost, and security requirements, optimizing the allocation of resources across the country.
Strategic Deployment of Large-Scale AI Clusters
The plan sets clear, ambitious targets for infrastructure deployment. It calls for the orderly establishment of intelligent computing clusters at a massive scale—ranging from 10,000 to over 100,000 accelerator cards. These clusters are essential for training cutting-edge AI models. Alongside this, the plan stresses the need to deploy inference computing facilities based on real-world application scenarios, ensuring that raw training power translates effectively into industrial and commercial use.
Prioritizing Domestic Chip Ecosystem Integration
A significant and highlighted directive within the plan is to "increase efforts to adapt and integrate domestic computing chips." This move signals a critical shift from proving the viability of homegrown chips to actively fostering their large-scale adoption within the most demanding computing environments. Large-scale cluster deployment provides the essential testing ground for optimization and validation, accelerating the journey toward a more self-reliant computing supply chain.
Beyond Hardware: Efficiency and Future-Proofing
The blueprint also addresses the "soft" capabilities required to maximize the value of computing investments.
Optimizing Resource Utilization
To tackle potential inefficiencies and mismatches between supply and demand, the plan advocates for deepening automated monitoring and supply-demand matching for computing power. The goal is to improve the overall utilization rate of national computing resources, preventing waste and redundant construction.
Investing in Next-Generation Research
On the R&D front, the plan takes a forward-looking stance. It emphasizes strengthening core and common technologies in computing. Notably, it also proposes to explore prospective research and ecosystem development for space-based computing technologies, a visionary nod to potential future paradigms. Furthermore, the plan commits to refining standards, evaluation systems, and continuing initiatives like the "Computing Power Foundation Strengthening" program to ensure robust and standardized industry growth.