On August 23, the 2025 Computing-Power Development Report was released during the opening ceremony of the 2025 China Computing-Power Conference.
According to the report, by the end of June this year China's AI-compute capacity had reached 788 EFLOPS (FP16), while infrastructure construction is simultaneously gaining both quality and speed.
The report notes that, in 2025, China is systematically building out a unified national computing-power network under a "point–chain–network–surface" framework. By optimizing the geographic layout of compute resources, strengthening collaborative technology innovation, deploying network infrastructure ahead of demand, and expanding application scenarios, the country is steadily raising the integrated supply capacity of its compute-and-network fabric.
As AI and other next-generation IT technologies leap forward, computing power has become a key driver of scientific innovation, industrial upgrading and social progress.
The report highlights that the current computing-power industry is expanding in a steady, well-structured manner and displays four overarching characteristics:
First, global computing-power capacity is growing steadily, with infrastructure undergoing continuous iterative upgrades.
Major economies are deepening policies in critical areas and accelerating infrastructure deployments. By the end of 2024:
- General-purpose compute (FP32) reached 628 EFLOPS worldwide
- AI compute (FP16) hit 5,693 EFLOPS
- HPC/supercomputing (FP64) totaled 20 EFLOPS
Computing resources are rapidly concentrating in the world's leading economies, while infrastructure is evolving toward higher performance and tighter coordination.
Second, top-level planning and design are being continuously refined, and demand for AI compute is surging.
At the national level, ministries are coordinating infrastructure planning to shift the computing-power industry from pure scale expansion to high-quality development.
At the local level, governments are tailoring resource layouts to their own conditions, creating a multi-tiered, differentiated regional computing-power development system.
At the industry level, the AI revolution is propelling explosive demand for computing power, with inference workloads poised to become the core driver of AI-compute capacity growth.
Third, infrastructure construction is accelerating in both quality and speed, while intelligent applications are emerging at a faster pace.
- Scale: By the end of June, China's live data-centre inventory had reached 10.85 million standard racks, delivering 788 EFLOPS of AI compute (FP16) and keeping supply and demand in dynamic equilibrium.
- Architecture: By the close of 2024, 50.3% of these facilities were directly connected to the national backbone, driving a step-change in network quality.
- Regional synergy: Also by June, the eight national computing-hub provinces housed 7.882 million standard racks-72.6% of the national total-widening both the throughput and the reach of each hub.
- Green performance: The average power-usage effectiveness (PUE) of China's live data centres fell to 1.46; more than 130 centres have now earned the CAICT 4A (or higher) green certification, cementing rapid progress toward low-carbon operations.
- Industrial uptake: With compute-network infrastructure upgraded and 10,000-GPU clusters being rolled out at record speed, generative-AI applications are springing up faster than ever.
Fourth, convergent innovation in the core segments of computing power is now proceeding bottom-up, with every layer driving the next.
- Compute technology: Rapid tech refreshes and home-grown hardware upgrades are shortening time-to-deployment, letting domestic accelerators scale from pilot lines to full-scale industry.
- Compute infrastructure: AI's own evolution is forcing continual optimisation-pooled scheduling, heterogeneous compatibility, elastic scaling and built-in security are now standard features.
- Compute platforms: One-stop marketplaces balance supply and demand in real time, turning raw FLOPS into finished AI services and giving enterprise users a guaranteed on-ramp.
- Compute applications: Foundation models and domain-specific agents are sinking deep into verticals, forging new productivity poles and adding fresh momentum to China's economic growth.
The report offers a comprehensive analysis of the global computing-power industry, systematically summarizes China's construction achievements, probes construction pathways and future trends in key segments, and-by pinpointing potential domestic bottlenecks-proposes high-quality development recommendations that provide a scientific roadmap for raising the technical level and overall efficiency of Chinese data centres.