China's AI Is Leaving the Lab. The Factory Floor Is Next.

China's AI Is Leaving the Lab. The Factory Floor Is Next.

For two days in Beijing last week, the world's most senior business figures gathered at the China Development Forum 2026 to hear a message that China has been refining for years: artificial intelligence is no longer a technology story. It is an economy story. And Beijing intends to write that story on its own terms.

The annual forum, held March 22 and 23 under the theme "advancing high-quality development and creating new opportunities together," marked a turning point in how China presents its AI ambitions. The focus had shifted decisively from frontier model capability to real-world deployment: AI in factories, on production lines, in security systems, embedded into the supply chains that underpin the country's industrial dominance.

From "AI+" to Intelligent Economy

China's "AI+" initiative has appeared in government work reports for three consecutive years, but this year's forum introduced sharper language. Officials unveiled the concept of a "new form of intelligent economy," a framework designed to move AI from national aspiration to measurable economic output.

Liu Liehong, a member of the Communist Party of China Leadership Group at the National Development and Reform Commission and head of the National Data Administration, gave the clearest signal yet of the scale Beijing envisions. He told forum delegates that by the end of the 15th Five-Year Plan period in 2030, China's AI-related industries are estimated to exceed 10 trillion yuan, roughly $1.45 trillion USD.

That figure is not a projection from an investment bank. It is an official benchmark from the body overseeing China's national data strategy. Liu also outlined a structural shift in how AI models are evolving: as foundation models mature, the emphasis is moving to domain-specific models trained on high-quality industry datasets, replacing the general corpora that powered the first wave of large language models.

As AI evolves from general-purpose foundation models to specialised industry-specific models, its integration with the real economy is becoming increasingly deep." — Liu Liehong, Head, National Data Administration of China

What the Factory Floor Actually Looks Like

The forum's AI industrial applications symposium drew delegates from across the global business community, and the examples on display were no longer aspirational. China's industrial parks are deploying AI-powered humanoid robots for production line tasks and security patrols. Facilities in Wuhan are generating 100 hours of daily training data for embodied intelligence systems, feeding humanoid robots the sensory and situational data they need to function reliably in unstructured environments.

Alibaba chairman Joe Tsai identified three pillars behind China's AI industrial readiness: heavy investment in the power grid, a commitment to open-source model development, and the country's manufacturing supply chain depth. These are not software advantages. They are physical infrastructure advantages that took decades to build and cannot be replicated quickly by competitors.

Apple CEO Tim Cook, also present at the forum, praised the sophistication of Chinese developers and the degree of automation visible in Chinese manufacturing facilities. Whether or not Cook's praise carries strategic intent, the observation itself is significant: the company that defined the smartphone era is watching China's AI-enabled factories closely.

"We'll See More DeepSeek Moments"

The most candid assessment from the forum floor came from a foreign voice. Denis Depoux, global managing director of Germany's Roland Berger strategy consultancy, told delegates he expected the pattern set by DeepSeek's January 2025 market shock to repeat itself.

I would assume that we will see more DeepSeek moments. AI development in China is very vibrant, with many models competing, and pretty much every month there is something new." — Denis Depoux, Global Managing Director, Roland Berger

Depoux's framing matters because it reframes the AI competition from a binary US-China frontier model race to something more complex: a continuous wave of application-layer breakthroughs, each one potentially reshaping competitive dynamics in specific industries. For foreign companies operating in or supplying to Chinese industrial markets, this is not a theoretical concern. It is a quarterly operating risk.

ByteDance has added to this dynamic with its Seedance 2.0 text-to-video model, which generates high-quality, cinematic-quality video clips from text prompts. The model represents the kind of creative AI application that disrupts not only content industries but also industrial design, marketing, and product visualisation workflows, areas where China's manufacturing base intersects with its growing consumer economy.

What This Means for the Region

China's industrial AI push has direct implications for Asia's broader technology landscape. Vietnam, which fired the starting gun on Southeast Asia's first AI law earlier this year, is watching Beijing's industrial AI model with interest. Southeast Asian manufacturers integrated into Chinese supply chains will increasingly encounter AI-augmented production systems as baseline expectations.

The Boao Forum in March estimated Asia's AI market at $400 billion by 2030, a number that looks conservative in light of China's own $1.45 trillion target for its domestic industry alone. The implication is that Asia's regional AI market projections systematically undercount Chinese domestic growth.

Singapore's position as Asia's per-capita AI leader highlights a different model: adoption-led rather than production-led. The tension between China's production-first approach and Singapore's adoption-first approach will define the region's AI competitive dynamics through the rest of the decade.

Events like SuperAI 2026 have brought these competing visions into dialogue, but the development forum made clear that Beijing is not waiting for international consensus. The factory floor is already being rewired.