Two Numbers, One Inflection Point

On October 19, China's National Bureau of Statistics presented its first-three-quarters industrial report card, containing two AI-relevant figures: AI adoption in key scenarios has exceeded one-third of industrial enterprises above designated size in manufacturing, and industrial robot output grew 29.1% year over year.

The same briefing signaled that the "AI Plus" initiative keeps releasing new economic momentum across manufacturing, services and information transmission — media samples even included "lights-out factories." The Ministry of Industry and Information Technology added it will accelerate the adoption of AI PCs and AI phones.

Key data at a glance

· AI adoption in key scenarios, large manufacturers: above 1/3 (NBS)

· Industrial robot output, first three quarters: +29.1% YoY (NBS)

· Reference point: Siemens plans to lift factory automation to 80% with AI (Investing.com)

What One-Third Actually Means

The metric counts adoption in key scenarios — not merely having AI software installed, but actually running it in quality inspection, production scheduling and equipment maintenance. Once more than a third of large manufacturers cross this line, AI's status in manufacturing shifts from "a pilot technology" to "a routine tool in the production mainstream."

The global race is accelerating on the same thesis. Siemens plans to raise factory automation to 80% with AI, per Investing.com; a Hankyoreh field report depicted a Hyundai plant where AI plans the paths of 1,700 welding spots. China's adoption data lands squarely in a window where major manufacturing powers are speeding up together.

From 'Have It' to 'Does It Work'

Once adoption passes one-third, the next competitive variable is depth per scenario: inspection moving from sampling to full coverage, scheduling from advisory to closed-loop execution, maintenance from post-mortem alerts to prediction. That demands deeper understanding of shop-floor data, process knowledge and safety constraints.

NineZenith's practice with manufacturing clients confirms the same trend: the bottleneck of LLM adoption is never inference itself, but engineering integration with OT data and production workflows — precisely the main battlefield of the application-led era.

(Information synthesized from public reports of the NBS briefing)