China's 'AI+Manufacturing' Push Tips Smart Factories Toward Scale

China is starting to talk about manufacturing and artificial intelligence as a single system rather than two industries. The clearest signal came in January 2026, when eight ministries led by the Ministry of Industry and Information Technology issued the action plan for "AI + Manufacturing", setting two practical targets: industry-wide foundation models tailored to each sector by 2027, and five hundred typical application scenarios promoted across the economy. The plan is the sector-level realization of the broader "AI+" initiative launched in August 2025, whose public objective is to lift AI penetration of production and management above 70 percent by 2027 and 90 percent by 2030. Behind the framing is a statistical story: penetration of AI in production and manufacturing process steps rose from 19.9 percent in 2024 to 25.9 percent in 2025, per Xinhua's reading of official data.

A base of factories, not just pilots

The strategy stands on a large installed base. The Ministry of Industry and Information Technology reported in February 2025 a three-tier national smart-factory system: more than 30,000 basic-level smart factories, roughly 1,200 advanced-level ones, and about 230 "excellence-level" plants, spread across all 31 provincial regions and covering more than 80 percent of manufacturing categories. For the excellence tier, the ministry's own figures claim average development-cycle reductions of 28.4 percent, productivity gains of 22.3 percent, product defect rates down 50.2 percent and carbon emissions down 20.4 percent. Those numbers are administrator-reported, directional rather than third-party audited — a caveat worth keeping in mind when this topic is quantified.

The World Economic Forum's Global Lighthouse Network offers a partly external check. Roughly 101 lighthouse factories sit in China, about 70 percent of the global cohort, including 16 new Chinese sites inducted in January 2026. Recent additions show the technology mix: EVE Energy's cylindrical-battery base was named the world's first cylindrical-battery lighthouse factory in January 2026; NIO's F2 plant in Hefei joined this June, running a digital-twin platform that manages more than 3.6 million vehicle configurations, with reportedly 90 percent of R&D workflows automated and AI-driven decision-making covering about 80 percent of its manufacturing scenarios. A digital twin coupled to battery-swap networks and in-vehicle AI, as NIO frames it, is a good illustration of where the boundary between "smart factory" and "robot fleet operator" is blurring.

Robotics: the dense underbelly

The most concrete part of the story is hardware. The International Federation of Robotics reports that in 2024 China installed 295,000 industrial robots, 54 percent of everything installed worldwide that year, and that its operating stock of roughly 2 million units is about 4.5 times Japan's, the second-largest fleet. Density, at 166 robots per 10,000 employees, still ranks only 22nd globally, a reminder that China's enormous manufacturing workforce spreads the fleet thin. The supply side has shifted: domestic suppliers outsold their foreign counterparts in China for the first time in this cycle, and domestic production now accounts for roughly 57 percent of Chinese demand, up from about a quarter a decade ago. The two robots that matter most for the future are harder to count: by late 2025 Chinese companies had released more than three hundred humanoid-robot products, and Chinese firms such as Unitree made up the majority of the estimated 16,000 humanoids sold globally in 2025. Physical AI — vision-language-action models that run directly on the factory floor — is the layer that turns a robotics story into an AI story.

Policy, subsidies, and the widening gap

Chinese policy is doing more than writing documents. Municipal and provincial robotics funds total well into the tens of billions of renminbi, and state-linked buyers are placing orders; China Mobile, for one, bought humanoid robots from domestic makers in 2025. This is the demand-side half of an old playbook that already rebuilt industrial robotics: roughly three quarters of China's robot demand was met by imports a decade ago; today the industry exports a rising share of global shipments. But the same subsidy machinery introduces distortion. The National Development and Reform Commission itself has warned that over a hundred robotics ventures are chasing very few proven, at-scale uses — a bubble risk the ministry's own data partly reflects, since figures are often self-reported in an incentive-heavy environment.

Analysis: what the scale-up does and does not mean

The significance of 2026 is that the direction has changed from penetration to scale. A regime that 30 percent of large-enterprise manufacturers say they already use, the share the MIIT cited in early 2026, is no longer a pilot program; it is the operating system of a portion of China's factory floor, and the policy target of 90 percent by 2030 is a commitment to make that portion the majority. Two implications follow that are not yet settled fact. First, the bottleneck has moved from access to generalization: the hardest problems are no longer a single smart line but moving physical-AI models across heterogeneous plants, each with its own machinery, layout and data — the same fragmentation seen in other domains where a model that performs in one plant frequently fails in another. Second, the export of robot hardware plus its embedded software is quietly becoming an industrial-policy lever: China's share of global industrial-robot exports rose from 5.9 percent in 2020 to 16.7 percent in 2024, and "dark" factories that capture value-added while employing little local labor are a growing concern for economies importing Chinese automation. None of that makes the numbers false; it makes the honest question the gap between administrator-claimed metrics and independently verified outcomes — the difference between a lighthouse certified by the WEF and the undifferentiated bulk of 30,000 plants behind it.