China computing power: How fast is capacity scaling, and where will future efforts focus?

Key Points

China computing power is heating up and the country is racing to turn raw capacity into practical, industrial-grade intelligence.

Overview — the current momentum

At the 2025 World Artificial Intelligence Conference, Huawei (Huawei 华为) showcased its Ascend 384 super-node system with a total computing capacity of 300 petaflops (PFLOPS).

At the 2025 China Computing Power Conference, the national computing platform reported comprehensive interconnection, signalling a national-level computing “brain.”

New versions of tools like DeepSeek are being adapted to next-generation domestic chips, showing software stacks are aligning with hardware advances.

The State Council has issued Opinions on Deepening the Implementation of the “AI+” Action, with a list of tasks focused on building nationwide computing capacity.

In the digital-economy era, computing power is now treated as a basic resource on par with water and electricity.

Key National Computing Power Metrics (as of June 2025)

Metric	Value	Context/Significance
Standard Racks	~10.85 million	Total active computing center racks, ranking second globally.
Intelligent Compute Scale	~788 EFLOPS	Leading segment driven by AI demand, indicating a “smart compute leading” pattern.
Total Storage Capacity	>1,680 EB	Significant storage for vast data generated by economic and social development.
Trunk Fiber Routes (East Data West Computing)	>250 planned/under construction	Enhancing data transmission efficiency between eastern and western regions.

Foundations are being solidified — scale, networks, and chips

“Scaling up” is the operational word for regional computing centers across China this year.

• China Telecom (Zhongguo Dianxin 中国电信) expanded its Central China intelligent computing center in Wuhan Future Science City by adding 300 intelligent compute servers, bringing that site’s total above 2,000 PFLOPS.
• The “Dong Shu Xi Suan” (东数西算, East Data West Computing) initiative’s Gansu hub in Qingyang now hosts a data-center cluster with about 31,000 racks, serving major regions including Jing-Jin-Ji (京津冀), the Yangtze River Delta, and the Guangdong–Hong Kong–Macau Greater Bay Area.

The China Academy of Information and Communications Technology (CAICT, Zhongguo Xinxi Tongxin Yanjiuyuan 中国信息通信研究院) finds the industry is optimizing structures while innovating technologically.

By the end of June, China had roughly 10.85 million standard racks in active computing centers, ranking second globally by total computing capacity.

For transmission, more than 250 “East Data West Computing” trunk fiber routes are planned or under construction.

On storage, total capacity has climbed above 1,680 exabytes (EB).

Computing comes in forms — supercomputing, general-purpose, and intelligent compute — and intelligent compute is the core foundation for AI.

Driven by AI demand, China’s intelligent compute scale has reached about 788 exaFLOPS (EFLOPS), making the pattern of “smart compute leading, with diversified coordinated development” visible.

On the tech front, ZTE (ZTE 中兴通讯) released an intelligent-compute super-node using its self-developed “Lingyun” AI switching chip and an open high-speed interconnect architecture.

That design lowers hardware barriers and lets domestic GPUs be flexibly combined for large-scale, high-speed collaboration.

Chips, software, and intelligent compute centers are moving from “usable” to well-tuned and practical.

Nationwide coordination is also maturing — in Hangzhou, Zhejiang, a 1 ms all-optical intelligent-compute private network enabled Leapmotor (Leapmotor 零跑汽车) to schedule compute resources efficiently.

Leapmotor reports manufacturing cycle time shortened from 60 months to 24 months thanks to compute–network collaboration.

A tiered national architecture is emerging: hub nodes, regional centers, and local edge sites.

The 2025 Computing Capacity Development Report shows early formation of latency tiers: 1 ms urban network, 5 ms regional, and 20 ms cross-national hub network.

Penetration into industry is deepening — real-world outcomes

Computing power is reshaping operations and innovation across sectors from mining to healthcare to automotive.

Mining example: At Xishan Coal Electricity Malan Mine (Shanxi Coking Coal Group, Shanxi Jiaomei 山西焦煤), dispatch operators can start remote cutting machines by pressing a button.

That shift has increased mining output per shift by more than 60% compared with traditional methods, enabled by intelligent systems running on modern compute infrastructure.

Economic multiplier: Industry estimates in the reporting suggest every ¥1 RMB invested in computing power can generate ¥3–4 RMB of GDP.

That means roughly ¥1 RMB ($0.14 USD) can drive ¥3–4 RMB ($0.42–$0.56 USD) in GDP gains, using an exchange rate near ¥7.2 = $1.

Automotive example: At Geely Xingrui’s (Geely Automobile 吉利汽车) intelligent compute center, systems handle requests and real-time regional data.

The center’s computing capability on the order of 102 × 10^18 operations per second supports millions of online vehicles needing real-time intelligence.

Geely’s research institute reports intelligent compute platforms can shorten model-training cycles for assisted-driving algorithms by over six months.

Healthcare example: Medical data in China is expanding rapidly, about a 36% compound annual growth rate reported in the article, and that creates explosive compute demand.

Guangdong Provincial Health Commission’s “Yueyi Zhiying” (粤医智影) uses compute and optimized algorithms to read imaging volumes equivalent to the daily work of 150 radiologists at a top-tier hospital every hour.

Tianjin Medical University General Hospital uses DeepSeek “intelligence-compute appliances” to provide customized compute services for elderly chronic-disease assessments and automated AI medical reports.

The Ministry of Industry and Information Technology (MIIT, Gongye he Xinxi Hua Bu 工业和信息化部) reports intelligent compute is widely used in large generative models, embodied intelligence, smart cities, and industrial manufacturing.

An industry-wide compute application competition collected over 23,000 innovative compute projects, many replicated at scale across finance, healthcare, energy, and manufacturing.

MIIT plans targeted actions to strengthen compute fundamentals — focusing on compute, storage, and networking — and to accelerate application of new technologies and products.

The ministry will deepen compute-enabled applications with special initiatives for education, healthcare, energy, and other priority sectors.

Industry Applications & Real-World Impact of Computing Power

Industry/Sector	Example/Impact	Key Metric/Benefit
Mining	Xishan Coal Electricity Malan Mine uses intelligent systems for remote cutting machines.	Mining output per shift increased by >60%.
Economic Multiplier	Investment in computing power stimulates GDP growth.	Every ¥1 RMB invested generates ¥3–4 RMB of GDP.
Automotive	Geely Xingrui’s intelligent compute center for real-time data handling and model training.	Shortened model-training cycles for assisted-driving algorithms by over six months.
Healthcare	Guangdong Provincial Health Commission’s “Yueyi Zhiying” and Tianjin Medical University General Hospital.	Healthcare data growth at 36% CAGR; imaging volumes equivalent to 150 radiologists/hour.

Where should future efforts focus? — priorities and technical gaps

China’s computing-power industry is moving into a phase of higher-quality, large-scale development, but challenges remain.

Key issues include supply–demand mismatches, gaps in critical technologies, and coordination problems between networks and compute.

Forecasts cited in the reporting suggest AI could contribute more than ¥11万亿元人民币 (¥11 trillion RMB, approximately $1.53 trillion USD) to China’s GDP by 2035.

That scale of growth could multiply demand for computing power by an order of magnitude or more, so scaling compute supply in a balanced, orderly way is essential.

A senior National Information Center official said the top priority for forward-looking digital infrastructure investment is intelligent compute centers.

Construction of smart compute centers should be guided by advanced system architecture and treat compute as infrastructure from civil engineering and hardware to full-stack deployment.

MIIT emphasizes orderly guidance for compute-facility construction to improve supply quality and help local authorities plan intelligent compute facilities rationally.

Interconnection and interoperability are vital to raising utilization efficiency.

The May-issued Action Plan for Compute Interconnection aims for basic nationwide standardized interconnection of public compute capacity by 2028.

The plan envisions a “compute internet” with intelligent sensing, real-time discovery, and on-demand allocation.

Experts advocate accelerating integration of the national compute network, strengthening unified resource coordination, and dynamic optimization.

They call for a coherent standard system for compute interconnection that covers resource sensing, task and data flow management, and application-architecture adaptation to build a unified market for compute services.

Full-chain innovation is critical.

China Mobile (Zhongguo Yidong 中国移动) executives suggest focusing on breakthroughs in:

• Ultra-large-scale intelligent compute clusters (100,000+ cards).
• Distributed inference architectures.
• Integrated compute–storage systems.
• Quantum and optical computing as potential disruptive technologies.

They also urge active participation in domestic and international standards bodies and open-source communities to export China’s compute-network solutions and foster a resilient ecosystem.

• Intelligent Compute Centers: Top investment priority for digital infrastructure, requiring advanced system architecture and full-stack deployment.
• Interconnection & Interoperability: Target basic nationwide standardized interconnection by 2028 for public compute capacity, forming a “compute internet.”
• Standard System for Compute Interconnection: Need for coherent standards covering resource sensing, task/data flow, and application adaptation.
• Full-Chain Innovation & Technical Breakthroughs: Focus on ultra-large-scale intelligent compute clusters (100,000+ cards), distributed inference, integrated compute–storage systems, and exploration of quantum & optical computing.
• Ecosystem Development: Active participation in domestic and international standards bodies and open-source communities.

Actionable takeaways for investors, founders, and builders

• Invest in intelligent compute centers — they’re the backbone of future digital infrastructure and a policy focal point.
• Prioritize interoperability — projects that enable cross-data-center task routing, resource sensing, and billing will be in demand.
• Edge + hub architectures win — latency tiers (1 ms / 5 ms / 20 ms) show where to place services depending on real-time requirements.
• Verticalized compute services — healthcare imaging, auto-AD model training, and industrial digital twins are ready markets.
• Full-stack value capture — from civil design to software orchestration, end-to-end integration yields higher utilization and faster ROI.

Conclusion

China is building a national compute backbone that mixes massive scale, regional optimization, and intelligent networking.

Numbers like 300 PFLOPS for Huawei’s Ascend 384, 10.85 million racks nationwide, 788 EFLOPS of intelligent compute, and 1,680 EB of storage signal both maturity and rapid growth.

To turn capacity into economic value, the priorities are clear: interconnection, standards, and full-chain innovation.

For investors, founders, and technologists, the opportunity is to build the software, orchestration, and application layers that unlock that compute for industry.

China computing power will be the backbone of the next digital economy wave.

References

• 算力热度持续攀升，如何“扩容”？ – 新华网
• 算力互联互通行动计划 / 相关政策与发布 – 中华人民共和国工业和信息化部
• Huawei Ascend and intelligent computing announcements – Huawei