AI, Data, and Compute
in China's 15th Five-Year Plan

China’s 15th Five-Year Plan treats artificial intelligence as a national production system, not just a technology race between model labs. That distinction is the key to understanding the document’s AI sections.

The plan does talk about model architecture, algorithms, multimodal systems, agents, embodied intelligence, and the path toward general AI. But it places those topics inside a larger stack: compute facilities, data resources, green power, industrial scenarios, digital services, governance rules, and international digital cooperation. In the plan’s logic, AI becomes infrastructure when it can be supplied, priced, governed, embedded, and used across the economy.

The short answer

China’s 15th Five-Year Plan makes AI part of the “Digital China” agenda. It focuses on three core inputs: compute, algorithms, and data. Then it pushes those inputs into industry, public services, consumption, education, healthcare, elderly care, government governance, and international digital cooperation.

The plan’s AI strategy has five layers:

Layer	What the plan wants
Compute	More national compute clusters, intelligent computing resources, cloud-edge-device coordination, green power integration, and lower compute costs for smaller firms.
Models and algorithms	Better model architectures, more efficient training and inference, multimodal AI, agents, embodied intelligence, industry models, and evaluation systems.
Data resources	National data accounting, public data sharing, authorized operations, enterprise and industry data development, AI corpora, high-quality datasets, and trusted data spaces.
AI Plus applications	AI applied across manufacturing, services, agriculture, education, healthcare, culture, social governance, consumption, and public services.
Governance	Data rights, circulation rules, revenue distribution, data security, AI regulation, international AI governance, and cross-border data cooperation.

The plan’s message is not “China wants chatbots.” It is “China wants intelligence to become a general-purpose factor of production.”

AI is framed as infrastructure

A consumer-facing AI product can look light: a chat box, a voice assistant, a shopping agent, an office copilot. The plan looks underneath that surface.

AI requires:

• chips and accelerators;
• data centers and power;
• cooling and network connections;
• model training and inference systems;
• high-quality data and corpora;
• software toolchains;
• standards and evaluation;
• industrial and public-service scenarios;
• legal rules for data use and safety.

The 15th Five-Year Plan tries to organize all of those layers. That is why its AI policy belongs as much to infrastructure and industrial policy as to software policy.

This also connects with the broader Chinese AI market. DeepSeek’s efficiency strategy matters because lower inference cost makes AI easier to deploy. OpenClaw’s product framework matters because AI becomes valuable when it is attached to real environments, accounts, workflows, and decisions. The plan is the state-level version of the same idea: AI has to move from model capability into usable systems.

Compute: the physical base of AI policy

The plan’s compute section calls for a multi-level compute infrastructure system and a nationally integrated compute network. It also mentions the “East Data, West Computing” project, national compute hubs, intelligent computing resources, cloud-edge-device coordination, and market-based compute services.

Several details are worth noting.

First, the plan emphasizes scale and coordination. China does not want compute to be only a scattered collection of private data centers. It wants national monitoring, scheduling, access, and matching capacity.

Second, the plan mentions green power and compute together. This matters because AI data centers are electricity-intensive. If China wants AI at industrial scale, compute policy becomes power policy.

Third, the plan calls for compute to become more accessible and lower-cost for small and medium-sized enterprises. That suggests a concern that AI infrastructure could otherwise concentrate in a few giant firms, cloud platforms, or government-backed clusters.

Fourth, the plan uses self-control language around software and hardware ecosystems. Compute capacity is not only about buying accelerators. It is about whether chips, systems, compilers, frameworks, cloud services, and applications can work together under external pressure.

Models: efficiency matters as much as spectacle

The plan calls for breakthroughs in AI theory and core technologies, model architecture improvement, algorithm optimization, and “model-chip-cloud-application” coordination. It also names multimodal AI, agents, embodied intelligence, group intelligence, and exploration of general AI.

This is a broad list, but one phrase deserves attention: efficient training and inference.

For a national AI strategy, capability alone is not enough. A model that is impressive but expensive to run has limited industrial reach. A model that is slightly less glamorous but much cheaper can spread through factories, offices, hospitals, schools, public agencies, and consumer services.

That is why Chinese AI competition should not be read only through benchmark rankings. The strategic question is whether Chinese labs and infrastructure providers can lower the cost of useful intelligence while preserving enough capability for real tasks.

This is exactly the logic behind the DeepSeek discussion on this site: DeepSeek V4 is best read as an efficiency strategy, not just a product launch.

Data: the plan’s most political AI input

Compute can be counted. Models can be benchmarked. Data is harder.

The plan calls for a national data resource system, a statistical survey system for data resources, a national “one ledger” for data resources, public data sharing, public data opening and authorized operation, enterprise and industry data development, data standards, quality management, AI corpora, and high-quality datasets for energy, transport, manufacturing, education, health, finance, and other fields.

That is a very ambitious data agenda.

It shows that China sees data as an economic input, but data is not like coal, land, or steel. It carries privacy, security, commercial, administrative, and political issues. Public data may be valuable for AI, but it sits inside government systems. Enterprise data may be valuable, but firms worry about rights, competition, and compliance. Personal data can train or improve services, but it raises consent and safety concerns.

The plan tries to solve this through institutions: data rights, circulation rules, revenue distribution, security governance, data markets, trusted data spaces, and data infrastructure. Whether those institutions create real trust will matter more than the slogans.

AI Plus: application is the point

The plan uses “AI Plus” to describe AI integration across the economy and society. This is not a narrow consumer app strategy. It includes:

• manufacturing digital and intelligent transformation;
• industrial internet systems;
• service-sector digitalization;
• smart agriculture;
• education applications;
• medical and health management;
• elderly care and disability services;
• employment and consumption;
• cultural production;
• government governance and grassroots public services.

The practical question is where AI has enough context to work. In a factory, AI needs process data, equipment interfaces, quality-control loops, and engineers who can act on outputs. In a hospital, it needs clinical workflows, regulation, liability rules, and trust. In education, it needs curriculum, assessment, teacher use, and safety boundaries.

The plan’s application-first language is important because AI productivity gains will not come from model access alone. They come from process redesign.

Governance is part of the product

The plan calls for a healthy and orderly development ecology. That phrase covers data systems, AI regulation, data security, privacy, market rules, cross-border data flows, and international AI governance.

For foreign readers, this may look like a constraint on innovation. Sometimes it will be. But in the plan’s logic, governance is also necessary for adoption. Firms and public agencies will not share data or deploy AI deeply if rights, liability, security, and regulatory expectations are unclear.

The hard part is balance. Too little governance can create safety failures, data abuse, low trust, and fragmented standards. Too much governance can slow experimentation, raise compliance costs, and concentrate AI development in the largest firms and state-backed platforms.

China’s 2026-2030 AI trajectory will depend on where that balance lands.

International digital cooperation has a Global South angle

The plan also points outward. It mentions digital trade, e-commerce, digital payments, smart cities, offshore compute facilities, cross-border data-flow service infrastructure, AI governance, digital currency, privacy protection, cross-border law-enforcement cooperation, and AI capacity building for Global South countries.

This is not only a domestic policy agenda. China wants its digital infrastructure, platforms, standards, and governance ideas to travel. The likely fields are practical rather than abstract: smart-city systems, e-commerce, payments, cloud services, industrial platforms, logistics, public services, and AI deployment in countries that need cheaper digital infrastructure.

That international layer should be read together with China’s wider opening and Belt and Road language. Digital infrastructure is becoming part of foreign economic policy.

What to watch

The plan’s AI agenda will be tested by implementation. Watch:

• whether national compute clusters become efficient shared infrastructure or fragmented local projects;
• whether compute pricing falls enough for small firms to use AI meaningfully;
• whether Chinese model labs continue to improve inference efficiency;
• whether public data can be used without creating privacy and trust problems;
• whether industrial AI adoption produces measurable productivity gains;
• whether data markets create real transactions or mostly administrative platforms;
• whether AI governance supports innovation or narrows it to a few trusted actors;
• whether China’s AI exports focus on software, cloud infrastructure, smart-city systems, or embedded industrial tools.

FAQ

Does China’s 15th Five-Year Plan focus on AI?

Yes, but not only as a model race. It puts AI inside the broader Digital China agenda, with compute, algorithms, data, applications, governance, and international cooperation.

What is AI Plus?

AI Plus means applying AI across industries, public services, social governance, consumption, education, healthcare, culture, and everyday life. It is a deployment agenda, not just a model-development slogan.

Why does compute matter so much?

Compute determines who can train, fine-tune, and run AI systems at scale. It also links AI policy to chips, data centers, electricity, cooling, networks, and cloud services.

What is the biggest uncertainty?

The biggest uncertainty is whether China can turn AI infrastructure into productivity gains across real organizations, rather than building capacity that remains underused or concentrated in a few platforms.

Sources

• State Council national development plan archive
• National People’s Congress resolution approving the 15th Five-Year Plan
• China’s 15th Five-Year Plan, Explained