Xinhua think tank report charts four trends in AI-enabled industrial development

The Xinhua news agency's China Economic Information Service released a think tank report on Sept. 9 that identifies four trends shaping how artificial intelligence is applied in industry, presenting them as an opening for the transformation of China's traditional manufacturing base. The report, titled Intelligent Transformation: Toward a New Human Industrial Civilization, was released at the 2026 Global Industrial Internet Conference in Shenyang, where Xinhua correspondents Li Yujia and Zou Mingzhong filed the story. Its stated subject is the revitalization of Northeast China's industry.

The findings reached readers through a single wire story. Xinhua's copy went live on news.cn and xinhuanet.com two seconds apart at 22:12 on Sept. 9, then reappeared on Sept. 10 at 00:52 on china.com.cn, at 09:05 on Zhejiang Online, and at 10:18 on Qiushi. Those are the same text under different mastheads, so the four trends rest on one originating report rather than on independent reporting by separate outlets.

The first trend describes a move from scattered pilots to enterprise-level operations. Leading manufacturers no longer confine AI to optimizing a single process; they use it to rebuild the operating system that connects research and development, planning, production, quality, logistics and the supply chain. The supporting figures are that as of 2025, analytical AI accounted for about 62 percent of lighthouse factory solutions, while generative AI had climbed to roughly 23 percent. Those shares describe lighthouse factories, the sites held up as benchmarks, and measure the leading edge of adoption rather than the typical plant.

The second trend names physical AI, systems that can perceive, act on and change the physical world, as the next competitive focus. The report argues these systems carry more long-term productivity gain than digital-only tools and are a route to easing labor shortages. The third trend concerns architecture: industrial intelligence is moving from cloud-hosted large models toward systems that coordinate the cloud with on-site operations. The report places the binding constraint in data and semantics rather than in model availability, a shift from the framing of recent years, so the open questions are whether equipment data can be captured in real time and whether an output can be returned safely to a production line.

The fourth trend is divergence. Manufacturing firms' use of AI has risen, unevenly across industries, while the frontier investment that pushes model capability forward is concentrating among companies with the capital to fund it. The report presents this as a double split between broader use on one side and narrower control over the direction of the technology on the other.

Set against earlier commentary on industrial AI, the notable element is the assertion that adoption has left the pilot stage. Prior framings treated the question as whether a factory had a model at all. This report treats that question as settled and moves the difficulty downstream. It does not disclose how the per-industry divergence was measured, and no manufacturer named in the circulated text has responded to its conclusions.

The copies in circulation end mid-sentence, so the report's recommendations for the Northeast and any estimates attached to them are not visible. The indicators to watch are the generative AI share in the next lighthouse factory update, which would test whether the 23 percent figure keeps climbing, and whether later editions quantify how concentrated frontier investment has become.