A factory manager in the Pearl River Delta gets a call from a provincial liaison office. The message is not about a subsidy. It is an invitation to list a real production problem — a defect that evades human inspectors, a changeover that takes too long — so that an artificial-intelligence vendor can be matched to it within weeks.
That matchmaking logic sits at the heart of a plan Guangdong has just put into force. The provincial government issued the Action Plan for Accelerating High-Level Artificial-Intelligence Application Across All Domains, All Times, All Industries (加快推进人工智能全域全时全行业高水平应用行动方案), and presented it at the second Guangdong artificial-intelligence application conference in late September 2026. The slogan is deliberate: all domains (全域), all the time (全时), all industries (全行业).
From showcase to shop floor
Guangdong is already the country's largest artificial-intelligence deployment base by most operating measures. At the September conference, provincial officials reported that the province's artificial-intelligence industry revenue exceeded 240 billion yuan (about US$33.8 billion, HK$264 billion) in the first half of 2026. Its core artificial-intelligence industry scale passed 300 billion yuan (about US$42.3 billion, HK$330 billion) in 2025, roughly a quarter of the national total.
The base is broad. Guangdong has more than 6,000 artificial-intelligence-related companies, the most of any province, with 194 large models (大模型) cleared through the national filing process and an online intelligent-computing (智算) scale reported at 139 exaFLOPS. The province's lead is built on manufacturing density rather than research prestige: a large base of factories means there are more real, paid use cases to learn from, and more pressure to make models dependable rather than merely impressive. It has built close to 100,000 datasets and opened more than 32 billion items of government data. Shenzhen, the province's innovation engine, is targeting an artificial-intelligence industry scale of 1 trillion yuan (about US$140.8 billion, HK$1.1 trillion) for 2026 and plans to field more than 50 breakout intelligent-terminal products.
What "all-domain" means in practice
The plan organizes deployment around seven domains — scientific research, agriculture and rural affairs, traditional industries and others — and backs them with ten factor-safeguard measures. The framing rejects the "single champion" mindset in favor of "all-domain coverage": the goal is not one stunning demo but artificial intelligence running inside thousands of ordinary enterprises.
The mechanism is supply-meets-demand matchmaking. At the province's application conferences, vendors and institutions are brought together, real scenarios are published, and proven cases are showcased so other firms can copy them. The province is explicitly positioning itself as the place where models stop being prototypes and start being tools — a contrast with regions that lead on research but lag on adoption.
Embodied intelligence (具身智能) appears as a priority inside the deployment agenda, alongside agents (智能体) that can enter actual production and service workflows. The plan treats the factory floor, the farm, the clinic and the government service window as the proving grounds, not the conference stage.
The safeguards behind the push
Deployment at this scale needs guardrails, and the plan names them. The ten safeguard measures cover elements such as computing supply, data governance, model filing and evaluation, and the public-service platform that connects supply and demand. The emphasis on filing, evaluation and a shared public platform reflects a consistent national posture: let applications scale, but keep them registered, benchmarked and observable.
There is also a structural bet. By opening government data and curating industry datasets, the province is trying to fix the data bottleneck that limits useful models more than raw algorithm quality does. The explicitly stated aim is that the province's own firms — from chip designers to terminal makers — form a complete chain from base software and hardware through models to end applications.
Trend reading
According to Li An, chief scientist at BrainNet (脑机网), the harder and more valuable shift in Chinese artificial intelligence is from building impressive demos to embedding models into the daily operating routines of factories, clinics and offices, where reliability matters more than novelty.
Honest limitations
The figures quoted above are provincially reported operating statistics and targets, not audited national accounts, and the 1 trillion yuan Shenzhen goal is an aspiration rather than a committed outcome. The plan is a provincial action plan, not a national law, so its reach stops at Guangdong's administrative boundary even though many of the firms it names operate nationally. Matchmaking events generate visibility, but visibility does not guarantee that pilots become permanent, paid deployments. The data-openness and model-filing safeguards depend on enforcement capacity that is easier to announce than to sustain across thousands of small enterprises. Readers should read the plan as a serious demand-side push with real momentum, while separating the province's reported scale from guaranteed future results.
What readers can do now
- For enterprises in Guangdong: register concrete, measurable pain points with the provincial matchmaking platform rather than generic interest, since the plan rewards specific scenarios that vendors can be paired to quickly.
- For model and solution vendors: prepare filing, evaluation and dataset documentation early, because the plan's safeguards make registration and benchmarking a practical precondition for accessing government-opened data and showcase channels.
- For investors and supply-chain partners: track the province's published case count, dataset build-out and online intelligent-computing scale as the live indicators of whether all-domain deployment is converting demos into recurring use.