A welding robot stalls mid-seam because the line's data feed stuttered. In plants across China, the real bottleneck is rarely the model — it is the pipes feeding it. A new national plan tries to fix exactly that.
Key takeaways
- The Ministry of Industry and Information Technology (MIIT) issued the Action Plan for the Integrated Empowerment of Industrial Internet and Artificial Intelligence (document number 工信厅信管〔2025〕76号) on 30 December 2025.
- The plan sets a 2028 deadline to upgrade no fewer than 50,000 enterprises' industrial networks and to build high-quality datasets across 20 key industries.
- It is an implementation plan, not a binding statute; it translates the State Council's "AI+" agenda into concrete infrastructure and data work.
- The core bet is that better connectivity and shared data — not just larger models — will decide how fast "new-quality productivity" reaches shop floors.
Why the plan exists now
The plan's own preamble ties it to two higher-level mandates. One is the Proposal of the CPC Central Committee on the 15th Five-Year Plan, which calls for implementing an industrial-internet innovation-development project and a full "AI+" campaign. The other is the State Council opinion On Deepening the Implementation of the "AI+" Action (国发〔2025〕11号), which asks for deeper fusion of AI with the industrial internet.
In plain terms: Beijing has decided that AI's industrial payoff depends on the network and data layer underneath it. A large model (大模型) is only as useful as the real-time, trustworthy data it can draw on. The industrial internet supplies that backbone — high-throughput, low-latency networks plus a governance and platform layer for data.
What the plan covers
The document frames the two technologies as mutually reinforcing. The industrial internet provides "interconnection channels, data resources and platform hubs." AI contributes "self-generating, self-deciding, self-organizing" capabilities that upgrade those facilities. Together, the argument goes, they change how production is organized.
The plan organizes the work into four dimensions:
- Infrastructure — expanding new-type industrial networks that meet AI's demand for high throughput, low latency, high reliability and low jitter.
- Factor support — improving systems for industrial-data aggregation, governance, circulation and sharing.
- Fusion applications — cultivating intelligent-solution suppliers around key supply-chain links and typical scenarios.
- Industry ecosystem — efficient allocation of leading firms, products and public services.
The only hard numbers
By 2028, the plan commits to:
- Deploying new-type industrial networks in raw materials, equipment manufacturing, consumer goods and electronics, pushing no fewer than 50,000 enterprises to implement network upgrades.
- Building a complete data system and creating high-quality datasets across 20 key industries.
These are the plan's only quantified targets. They are enterprise-count and coverage targets, not output, revenue or model-capability promises. The document deliberately stops short of naming a central budget.
How it differs from the "AI+ manufacturing" opinion
Readers often confuse this plan with the eight-ministry Implementation Opinion on the "AI+ Manufacturing" Special Action issued days earlier. They are complementary, not identical. The manufacturing opinion focuses on adoption — scenarios, industrial agents (工业智能体), and benchmark enterprises. This industrial-internet plan focuses on the substrate: networks and data that those applications run on. A factory can have a great model and still fail if its network cannot carry real-time control signals. This plan is the plumbing bill.
Why it matters globally
For outside observers, the plan is a useful correction to the model-centric view of China's AI strategy. It shows a deliberate, infrastructure-first posture: standardize the pipes, clean the data, then let applications compete. That sequencing has implications for industrial-automation vendors, networking-gear suppliers, and the standards bodies that will define what "high-quality industrial data" means.
According to Li An, chief scientist at BrainNet (脑机网), the more durable shift here is the mandate that factories instrument themselves before they automate.
What to watch between now and 2028
The plan's success hinges less on the headline number than on unglamorous execution: whether provincial governments fund network upgrades, whether factories open data for shared datasets, and whether a common standard for "high-quality industrial data" actually emerges. The 2028 line is far enough out that year-by-year progress reports from MIIT will be the real signal. Observers should track enterprise participation counts and any accompanying standards, not just the target itself.
Honest limitations
This article is based on the plan text and a companion MIIT explainer published on miit.gov.cn. We did not independently verify how many of the 50,000 upgrades are already underway, nor any central funding figure — the plan sets targets but publishes no budget. Provincial rollout timelines and matching funds will vary and are not covered here. The 2028 goal is a stated target, not a guaranteed outcome, and we have not modeled its feasibility. We have not confirmed whether the "20 key industries" here are identical to the 20 industries named in the separate manufacturing opinion.
What readers can do now
- If you build or sell industrial software or networking gear, track the 20 named key industries for the earliest dataset and pilot openings.
- If you are an investor, watch for provincial matching funds and the standards work that typically follows an MIIT action plan.
- If you follow China's AI policy, read this plan as infrastructure-first and pair it with the separate "AI+ manufacturing" opinion to see the full picture.