China’s manufacturing base could give it an edge in industrial AI, Cisco executive David Siroky says, because factory lines, sensors and their operational data can train models to spot defects and predict equipment failures. The examples show why the AI race increasingly depends on the industrial infrastructure around the models, not just the models themselves.
Cisco Watch analysis
What happened
Siroky, Cisco’s AI architecture lead for Asia Pacific, Japan and Greater China, told China Daily that China’s manufacturing infrastructure could support physical AI. One example is BMW’s Shenyang paint shop, where an AI-assisted computer-vision system scans images of cars for scratches and marks repair areas with lasers.
The article also reports that China accounted for 59 per cent of industrial robots installed worldwide in 2025, citing the International Federation of Robotics. Turning factory data into material suitable for training models remains a separate task: Chinese government guidelines set a target of 100 high-quality industrial datasets by 2027.
Why it matters
Factory AI needs more than a capable model. It needs useful operational data, systems that can act on model outputs, and the computing and networking to move information around. Siroky says companies often discover during pilot projects that their networks cannot provide enough throughput, or that their facilities cannot power dense GPU clusters.
That creates a commercial opportunity for Cisco in high-speed networking. It also puts a practical question behind the industrial AI pitch: can companies reliably turn production data into systems that catch defects or predict failures, at a cost that makes deployment worthwhile?
Our read
The BMW example gives this story a concrete job for AI: find tiny paint defects early enough to repair them on the factory line, rather than after a car has travelled overseas. The broader adoption figures and Cisco’s commercial interest are useful context, but they do not establish that every factory is ready to follow suit. The unglamorous combination of clean data, reliable equipment and enough electricity may decide more than another model launch.
What to watch
- Whether the target of 100 industrial datasets by 2027 translates into usable training data.
- Whether factory AI systems show measurable gains in defect detection or equipment maintenance.
- Whether networking capacity and power constraints slow projects as they move beyond pilots.
Discussion spark: Should factories prioritise AI systems that catch costly defects on the production line, even if building the data and computing infrastructure is expensive?
Sources and evidence
- Cisco exec sees China holding sway in industrial AI – Chinadaily.com.cn (6 November 2026, 01:22 UTC)
Independent WittyWires tracker for public updates about Cisco. Not affiliated with or endorsed by Cisco; this is not an official account.