China’s government has issued a rapid sequence of policy directives aimed at building a nationally coordinated computing grid to support the country’s artificial intelligence ambitions and reduce reliance on foreign technology. The Ministry of Industry and Information Technology (MIIT), working with other departments, has laid out infrastructure targets, interoperability standards, and flexible commercial models that together form the scaffolding of what policy documents call a “national integrated computing power network.” The effort is a major state-led technology buildout, and it arrives as U.S. export controls tighten the supply of advanced chips to Chinese firms.
A State-Steered Compute Grid Takes Shape
The backbone of the strategy is a plan issued by the MIIT and five other departments titled the Computing Power Infrastructure High-Quality Development Action Plan, which sets national-level targets for computing infrastructure through 2025. The plan covers data center expansion, network upgrades, and the creation of standardized metrics and management frameworks, all intended to ensure that AI workloads can be trained and deployed on domestically coordinated resources rather than piecemeal commercial clusters. It also encourages the construction of large-scale, energy-efficient facilities and backbone networks that can support high-bandwidth, low-latency connections between major urban centers and remote server hubs.
Layered on top of that framework is a joint directive from the National Development and Reform Commission, the National Data Administration, and three other agencies that deepens what Beijing calls the “East Data, West Computing” initiative. That document, which details the construction of a nationwide integrated computing power network, includes tiered latency targets of 1ms, 5ms, and 20ms depending on the task, along with green power requirements and adoption concentration goals in designated hub regions. The idea is to route data-heavy eastern demand to cheaper, energy-rich western provinces while keeping response times tight enough for real-time AI inference. A separate State Council implementation plan stated that China should form a preliminary compute infrastructure system by end-2025, signaling that the leadership expects tangible national coverage rather than just pilot projects within the current five-year planning window.
Interoperability and Pay-by-the-Hour AI
Building raw capacity is only half the challenge. The MIIT also issued a separate directive, the Computing Power Interconnection and Interoperability Action Plan, which tackles the problem of making compute resources from different providers and regions work together seamlessly. The plan promotes high-speed interconnect standards, compute scheduling and dispatch protocols, and security requirements, all designed to let a startup in Shenzhen tap GPU clusters in Guizhou as easily as local servers. It envisions unified interfaces and common service catalogs so that users can request processing power, storage, or AI acceleration without needing to understand the underlying hardware or network topology.
One of the more commercially significant details in that plan is the introduction of flexible consumption models for AI workloads. The directive outlines commercialization pilots built around billing units described as “card-hours” and “machine-hours,” designed specifically for large-model training and inference. This pricing structure would let smaller firms and research labs buy compute in increments rather than committing to expensive long-term contracts, lowering the barrier to entry for domestic AI development. By standardizing how usage is measured and billed across different regions and providers, the state is attempting to create a quasi-utility model for AI compute, where capacity can be traded, dispatched, and priced under common rules that favor broader adoption over bespoke enterprise deals.
Engineering Standards Move from Strategy to Specs
The policy push has now advanced beyond broad strategy documents into granular technical specifications. The National Data Administration and the National Data Standardization Technical Committee recently solicited public comment on seven technical documents, including new compute grid-connection requirements for the national integrated network. These draft standards cover resource management and scheduling, multi-dimensional billing, efficiency measurement, operations services, and monitoring, the kind of engineering detail that turns a political vision into procurement rules and compliance benchmarks. They specify how different classes of computing power should be cataloged, how service quality is to be measured, and how operators must report utilization and energy efficiency to regulators.
Separately, the MIIT opened a public comment period on a Computing Power Standards System Construction Guide for 2025, which formalizes a national standards system aligned to the integrated network. The practical effect of these documents is that any data center operator, chip designer, or cloud provider wanting to participate in the national grid will eventually need to meet state-defined benchmarks for performance, billing transparency, and interoperability. That creates a domestic ecosystem with its own rules, potentially less dependent on foreign standards and supply chains, and gives regulators levers to steer adoption toward technologies that comply with Chinese specifications.
Self-Reliance Under Export Pressure
None of this is happening in a vacuum. Beijing’s compute buildout is directly shaped by the tightening of U.S. export controls on advanced semiconductors. As the Associated Press has reported, China’s latest economic planning emphasizes speeding up self-reliance in science and technology, a posture tied to political decisions around the country’s next development cycle and driven in part by external pressure from Washington’s chip restrictions. The compute grid is the infrastructure answer to that political directive: if China cannot freely buy the most advanced chips, it will try to maximize the output of whatever hardware it can produce or procure domestically, spreading workloads across a larger number of interconnected facilities to approximate the performance of more powerful but restricted processors.
China has also issued interim measures to both support and regulate generative AI applications, signaling that the government views the technology as a strategic asset that requires state guidance rather than unchecked commercial development. Those provisional rules on generative services set out requirements for security assessments, content oversight, and algorithmic transparency, effectively tying access to large-scale compute with obligations around how AI models are trained and deployed. In practice, the combination of AI service rules and a state-backed push to standardize computing infrastructure gives authorities additional levers over how large-scale AI systems are developed and deployed, while still encouraging domestic innovation.
From Infrastructure Blueprint to AI Industrial Policy
Taken together, the infrastructure plans, interoperability directives, and technical standards amount to more than a narrow technology program; they function as a broad industrial policy for AI-era computing. The MIIT-led push to define national targets for data centers and backbone networks ensures that capital-intensive construction is guided by state priorities rather than left solely to market demand. The “East Data, West Computing” framework overlays those investments with a geographic logic that attempts to balance energy use, land constraints, and industrial clustering, effectively turning remote provinces into back-end engines for coastal digital economies.
At the same time, the emphasis on standardized billing units and shared service catalogs is meant to keep the resulting capacity from being locked up by a handful of large cloud providers. By encouraging card-hour and machine-hour pricing, regulators hope to make high-performance computing accessible to universities, small and medium enterprises, and local governments that might otherwise be priced out of cutting-edge AI infrastructure. The emerging standards regime, with its detailed connection rules and performance metrics, gives Beijing the tools to enforce these goals over time, allowing it to nudge the market toward preferred technologies and business models while maintaining political control over the core infrastructure that will underpin the country’s next phase of digital development.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
