Beijing's Plan for Computing Power Infrastructure Construction Released, Mentioning Support for Procurement of Independently Controllable GPU Chips.

On April 24th, the Beijing Municipal Economic and Information Technology Bureau and the Beijing Municipal Communications Administration jointly issued the "Beijing Municipal Plan for Computing Power Infrastructure Construction (2024-2027)". In order to earnestly implement the decision-making and deployment of the Party Central Committee and the State Council on deepening the research and application of artificial intelligence, to effectively carry out the "AI+" action plan, to moderately advance the construction of digital infrastructure, and to accelerate the cultivation of the computing power industry ecosystem, according to the "Implementation Opinions on Deeply Implementing the 'East Data West Computing' Project to Accelerate the Construction of a National Integrated Computing Power Network" jointly issued by the National Development and Reform Commission and four other ministries and commissions, and the "Action Plan for High-Quality Development of Computing Power Infrastructure" jointly issued by the Ministry of Industry and Information Technology and five other ministries and commissions, combined with the actual situation of the city, this implementation plan was formulated.

Overall Approach

In response to the significant needs of economic and social development, scientific and technological innovation, and industrial transformation and upgrading, especially the intelligent computing power demands of the artificial intelligence industry development, the plan coordinates and improves the collaborative development layout of computing power in the Beijing-Tianjin-Hebei-Mongolia region, optimizes the structure of computing power supply, enhances the computing power scheduling capability, establishes a mechanism for the integrated and collaborative development of computing power in the Beijing-Tianjin-Hebei-Mongolia region, and constructs a computing power infrastructure foundation that integrates information computing power, network carrying capacity, and data storage capacity, aiming to build the Beijing-Tianjin-Hebei-Mongolia area into a highland of intelligent computing power innovation and application with high-quality supply, high-speed linkage, efficient scheduling, and high-level application.Demand-driven, regional collaboration. Accurately grasp the strategic positioning of the capital city, take market demand as the guide, actively integrate into the development of the East Data West Computing initiative, and build a closely collaborative and supply-demand coordinated Beijing-Tianjin-Hebei-Mongolia computing power integration and collaborative development pattern. Support the acceleration of computing power infrastructure construction in Tianjin, Hebei, Mongolia, and western regions, and prioritize the city's intelligent computing demand towards these areas; the city will no longer add general computing power in principle, and the insufficient part will be met by Tianjin, Hebei, Mongolia, and western regions.

Advertisement

Innovation-driven, autonomously controllable. Firmly adhere to the route of autonomously controllable computing power, strengthen the research and development of key software and hardware technologies such as AI chips, operating systems, and databases, support the development and breakthrough of new technological routes such as silicon photonic chips, quantum chips, and integrated storage and computing, and encourage the construction of computing power infrastructure to adopt safe and reliable software and hardware products. Gradually improve the level of autonomous control of core software and hardware in newly built and renovated intelligent computing centers, and provide favorable conditions for the continuous iteration of core software and hardware products.

Market-led, government-guided. Fully leverage the decisive role of the market in resource allocation and support various market entities to actively participate in the construction and operation of computing power infrastructure. Play the guiding role of the government to form a supply-demand linkage, open competition, and rational layout of computing power construction and operation mechanisms. Accelerate the construction of a domestic leading computing power interconnection and operation service platform, establish an intelligent computing resource pool that meets the needs of the city's AI companies, build an intelligent computing operation service platform through market mechanisms, and achieve the optimal allocation of intelligent computing resources.

Energy-saving and efficient, safe and reliable. Accelerate the layout and construction of key intelligent computing centers, improve the energy efficiency standards of computing power infrastructure, and use policy and market means to gradually guide the existing data centers to complete upgrades, transformations, or relocations. Encourage the transformation and upgrading of existing data centers into intelligent computing centers, support the green and energy-saving transformation of existing data centers, and promote the relocation and resolution of inefficient existing data centers. Comprehensively enhance the ability to resist risks, strengthen the security protection of key infrastructure and data security management, and ensure the safe and stable operation of important information systems.

---

Planning Objectives

By 2025, a pattern will be essentially established where intelligent computing resource supply is clustered, intelligent computing facility construction is autonomous, intelligent computing capability empowerment is precise, intelligent computing center operations are green, and the development of the intelligent computing ecosystem is systematic, leading the Beijing-Tianjin-Hebei-Mongolia region to become an internationally influential intelligent computing industry innovation and application highland. By 2027, the quality and scale of computing power supply in the Beijing-Tianjin-Hebei-Mongolia region will be optimized, striving to meet the training needs of large models with autonomously controllable computing power, the computing power energy consumption standards will reach a leading level in the country, and a rational layout, computing network collaboration, technologically advanced, green and intensive, and complete industrial chain of Beijing-Tianjin-Hebei-Mongolia computing power integration and collaborative development pattern will be established, integrating into a national integrated computing power system that is interconnected, widely accessible, green, and secure, providing strong support for the implementation of the national AI industry development strategy.

(1) Clustering of intelligent computing resource supply

Change the "small, scattered" situation of intelligent computing construction, and concentrate on building a number of large intelligent computing clusters. By 2025, the city's intelligent computing supply scale will reach 45EFLOPS, and from 2025 to 2027, the layout of computing power will be further optimized according to the development needs of AI large models and relevant national deployments. The functions of the Beijing computing power interconnection and operation service platform will be gradually improved, and the supply and demand scheduling of intelligent computing resources will be efficient and coordinated.

(2) Autonomous construction of intelligent computing facilitiesBy 2027, achieve full-stack autonomy and controllability of smart computing infrastructure software and hardware products, with overall performance reaching a leading level in the country, possessing 100% autonomous and controllable smart computing center construction capabilities, effectively supporting the training and inference of general and industry-specific large models that benchmark international leading levels.

(III) Precise Empowerment of Smart Computing Capabilities

Smart computing capabilities are precisely empowered to the city's governance, industrial development, and livelihood protection, with more than 200 industry benchmark cases driven by smart computing, promoting the digital and intelligent transformation in various fields such as healthcare, education, culture, finance, manufacturing, and energy.

(IV) Green Operation of Smart Computing Centers

The PUE value of newly built and renovated smart computing centers in the city generally does not exceed 1.25, and for large-scale advanced smart computing centers with annual energy consumption exceeding 30,000 tons of standard coal, the PUE value generally does not exceed 1.15. Promote the upgrade and transformation of existing data centers in the city, and by the end of the planning period, the PUE value of all existing data centers will not be higher than 1.35.

(V) Systematization of Smart Computing Ecosystem Development

A smart computing ecosystem with a complete industry chain, open cooperation, and integrated innovation is essentially established, with continuous improvement in smart computing chip design and production capabilities, increasing operational levels of smart computing cluster construction, a more open data factor market, and more efficient iterations of artificial intelligence technology, forming a "best practice standard system" that promotes the development of the smart computing ecosystem.

Key Tasks

(I) Promote Independent Innovation in Computing Power IndustryAccelerate the promotion of independently controllable core software and hardware products. Accelerate the layout of advanced computing, software tools, operating systems, and other key technologies of computing power, focusing on breaking through a batch of symbolic technical products and solutions. Build an independently controllable chip testing and verification platform, promote the improvement of the performance and reliability of independently controllable chips, accelerate the research and development and industrialization process of independently controllable chips, and speed up the construction of the independently controllable computing power technology system. Vigorously promote the adaptation of artificial intelligence large models with independently controllable chips, and improve the safety, stability, and resilience of China's intelligent computing industry supply chain.

Implement the intelligent computing center benchmark demonstration project. Build intelligent computing centers based on independently controllable artificial intelligence chips, training frameworks, and interactive networks, and accelerate the construction of integrated software and hardware computing power benchmark solutions and application ecosystems. Strengthen frontier exploration and forward-looking layout, and promote the development of advanced technologies such as storage-computing integrated chips, silicon photonic chips, quantum computing, optical interconnection, reconfigurable networks, and cloud-native. Encourage enterprises to explore the construction of intelligent computing centers using new technologies and new architectures such as optical interconnection and optical computing, and accelerate the application of new computing power technologies.

(II) Build an efficient computing power supply system

Build an urban computing power supply system dominated by intelligent computing power, with new generations of supercomputing, cloud computing, blockchain, and other multi-element collaborative systems. Focus on building Haidian, Chaoyang, Yizhuang, Jingxi (Shijingshan, Mentougou) and other E-level intelligent computing power highlands, and prioritize the acceleration of the construction progress of two 10EFLOPS large-scale intelligent computing clusters, focusing on meeting the rapidly growing demand for computing power for large model training and inference.

Establish a unified identity authentication, computing power resource scheduling, computing power resource service, and computing power transaction settlement platform for Beijing's computing power interconnection and operation services, and continue to expand the scale of intelligent computing resources in the Beijing-Tianjin-Hebei-Mongolia and western regions. Form a high-quality, efficient, and leading domestic intelligent computing operation service and resource allocation mechanism through market-oriented means, and strive to provide inclusive intelligent computing resources for large model enterprises and research institutions.

(III) Promote the integrated construction of Beijing-Tianjin-Hebei-Mongolia computing power

In accordance with the layout of the national integrated computing power network hub nodes, build an integrated and collaborative development pattern of Beijing-Tianjin-Hebei-Mongolia computing power, and create a Beijing-Tianjin-Hebei-Mongolia computing power supply corridor with Inner Mongolia (Holingol, Ulanqab) - Hebei (Zhangjiakou, Langfang) - Beijing - Tianjin (Wuqing) as the main axis. Support local enterprises in building computing power infrastructure in the above areas, plan and build large-scale intelligent computing clusters that support the training needs of trillion-level parameter large models, and gradually form a regional collaborative computing power supply system with a gradient distribution, reasonable layout, and complete functions.

Accelerate the improvement of computing power infrastructure carrying capacity. Promote the widespread coverage of large bandwidth, low latency all-optical access networks, and coordinate the construction of direct connection networks for key computing power centers, connecting the main computing power resources in the city, with network communication bandwidth reaching more than 400G and round-trip network latency controlled within 1 millisecond. Build a one-hop direct connection network from Beijing to the Beijing-Tianjin-Hebei hub according to demand, optimize the optical cable route, reduce detour latency, and promote the realization of round-trip network latency between key computing power centers within 200 kilometers of Beijing not higher than 3 milliseconds, and round-trip network latency between Beijing and western nodes such as Holingol, Inner Mongolia, not higher than 5 milliseconds.

(IV) Improve the green and low-carbon level of intelligent computing centers

Encourage existing data centers to upgrade to intelligent computing centers without increasing the total energy consumption, or use efficient system designs such as liquid cooling, modular power supplies, and modular machine rooms to reduce PUE and CUE indicators. For those that cannot complete energy-saving upgrades and have not transformed into intelligent computing centers, guide them to relocate and shut down. New and renovated intelligent computing centers increase the coverage of green energy-saving technologies and equipment, strengthen the use of green energy-saving measures such as photovoltaic power generation and waste heat recovery, and improve the energy and carbon efficiency of computing power infrastructure. Create benchmark demonstration zero-carbon intelligent computing centers that use 100% green electricity.(V) Deepen the Empowerment of Computational Power in Industry Applications

Strengthen the intelligent computational power support for the development of the artificial intelligence industry, accelerate the commercialization and landing of large AI models, promote the application and landing of large models in urban governance, government services, education and research, medical health, intelligent manufacturing, cultural tourism, financial technology, smart energy, and other fields. Support enterprises to delve into vertical fields to refine and strengthen industry models, build an efficient and collaborative large model innovation ecosystem, cultivate a group of influential large model service providers, and promote the integration and application of computational power to broader, deeper, and more refined areas, driving industrial upgrading and transformation.

(VI) Ensure the Safe and Stable Operation of Computational Power Infrastructure

Construct a comprehensive defense system for the security of computational power infrastructure to ensure the safe and stable operation of computational power infrastructure and key information systems. Enhance network security protection capabilities, carry out communication network security protection work, and strengthen security technical measures; improve data security protection capabilities, strengthen data classification and grading protection, and implement precise and strict management of important and core data according to regulatory requirements. Ensure the stable operation of computational power infrastructure, strengthen computational power network protection, adopt dual-node and dual-routing configurations for important network facilities to avoid single-point failures.

Guarantee Measures

(I) Strengthen Coordination

Establish a collaborative and operational guarantee mechanism for the construction of smart computing centers within Beijing, and a regional linkage mechanism among Beijing, Tianjin, Hebei, and Inner Mongolia. Coordinate the supply of smart computing chips to meet the construction needs of smart computing clusters, conduct centralized supply and demand docking, improve the efficiency of smart computing resource allocation, strengthen the coordination of smart computing resources with Tianjin, Hebei, Inner Mongolia, and the western regions, and ensure that the plan continues to effectively serve the development of artificial intelligence.

(II) Improve Policy Guarantees

Establish a coordination mechanism for energy indicators and green electricity supply to support the construction of smart computing centers, promote the adjustment of energy consumption structures within the industry or region, gradually expand the level of green electricity consumption in existing data centers, deepen support measures for the transformation and upgrading of existing data centers, and use reduced energy consumption to ensure the energy demand of new smart computing centers. Clarify the assessment standards for the use of green electricity in newly built and renovated smart computing centers to ensure the supply of green electricity and improve the energy usage structure of smart computing centers in the city.(III) Expand Financial Support

For enterprises that purchase independently controllable GPU chips to provide intelligent computing power services, support will be provided at a certain proportion of the investment amount to accelerate the realization of self-controlled supply of intelligent computing resources. For existing data centers that actively undergo green energy-saving transformations, support will be provided at a certain proportion of the investment amount. Enhance the policy effectiveness of artificial intelligence computing power vouchers, encourage enterprises to make good use of intelligent computing power resources, and accelerate the empowerment of large models in industry applications.

(IV) Deepen Exchange and Collaboration

Regularly organize forums, summits, and other special events to promote the collaborative supply and demand of intelligent computing power. Fully leverage the driving role of industry associations and industry alliances, and organize enterprises in Beijing, Tianjin, Hebei, and Inner Mongolia to engage in exchanges and cooperation in technology research and development, product promotion, talent training, and other aspects. Actively participate in the formulation of international rules, national standards, and industry norms to elevate the level and quality of international development. Utilize platforms such as the Zhongguancun Forum, China International Fair for Trade in Services, and the Global Digital Economy Conference to conduct industry discussions, exhibitions, and other activities, expanding the influence of the city's artificial intelligence industry development.