On 29 September 2026 the Policy and International Cooperation Institute of the China Internet Network Information Centre (CNNIC) released the Generative AI Application Development Report (2026) at the seventh China Internet Infrastructure Resources Conference. The figure that attracted most attention is that, by the first half of 2026, generative AI users in China had passed 700 million, a penetration rate above 50.0%. Beyond that figure the report discloses data on application scenarios, computing power, models, embodied intelligence and investment and financing. Read together, they reveal three structural changes — and each corresponds to a new legal and compliance position for AI companies.
Half the population using it: what passing the critical point means
Seven hundred million users and a penetration rate above 50% mean that generative AI has moved beyond the stage of early adoption and become a general service for the majority of the population.
From a legal perspective, “half the population” is not a marketing figure but a reference point for the scale of responsibility. Where a service reaches the majority of the population, several categories of obligation are enlarged at once:
On the content side, the requirements for labelling generated synthetic content and for handling misinformation and harmful information now operate in relation to the whole of society rather than to early adopters. On the service side, requirements for handling complaints and reports, staffing human customer service and providing a minors mode must be matched to the real user base. On the data side, once the number of individuals whose personal information is processed reaches the order of tens of millions, the undertaking may fall within the scope of the (draft) Provisions on Personal Information Protection by Large Personal Information Processors.
This linkage between user scale and compliance obligation is the report’s most direct message for companies. On the way in which the filing system tightens as the service becomes general, see the analysis of generative AI filing data for July and August 2026.
From question-answering tool to daily assistant: the migration of scenarios
The data on application scenarios show what users are doing with AI:
76.0% of users say they ask generative AI to answer questions — question answering remains the principal entry point; 47.8% process images or video, 37.6% process text, and 32.5% use it for work summaries, meeting minutes and presentations. More noteworthy is the change in frequency: use of AI general assistants and AI productivity tools grew by more than 100% year on year.
The migration from occasional questions to daily use brings new compliance touchpoints. Meeting minutes involve the information of others, office documents involve trade secrets, and image and video processing involves synthetic content labelling. These scenarios move AI from an information tool into the workflow, and correspondingly engage the rules on personal information processing, confidentiality and labelling of generated content. For an overview of the domestic rules on AI information services, see the survey of domestic AI regulation.
The report also notes the release of intelligent consumption: 38.7% of internet users had bought smart hardware in the previous six months, of whom 20.2% bought smart wearables and 18.2% bought 3C products such as phones and tablets. The opening of hardware entry points means that the compliance questions raised by on-device AI — local processing of data and filing of on-device models — will become more widespread as shipments grow. See the discussion in filing for on-device generative AI services.
Autonomy in the industrial chain: computing power, models and embodied intelligence
The report’s other thread is the progress of self-sufficiency in the industrial chain.
On the computing foundation, by June 2026 China’s intelligent computing power reached 2,185 EFLOPS, an increase of 177% year on year. On model capability, the report states that the top six models on the global ranking of model invocations all come from Chinese teams, and that domestic models and domestic computing chips are being adapted at pace. On embodied intelligence, the report identifies it as one of the most active directions in AI development: in the first half of 2026 Chinese manufacturers brought more than 400 humanoid robot products to market, and full-year output is expected to exceed 100,000 units.
These three sets of data point to the same conclusion: from computing power to models to hardware carriers, the key links in the industrial chain are developing self-sufficient supply capacity. For legal services, a complete chain means differentiated demand — the model layer is concerned with training data and open-source liability, the application layer with filing and content labelling, and the hardware layer with on-device data processing and product liability, while companies expanding overseas must also reconcile multiple jurisdictions. On the relevant capability, see multi-jurisdiction compliance.
The other side of investment: transactions and legal support
The final set of data concerns market activity: in the first half of 2026 there were 1,255 AI-related investment and financing events in China, already 78.0% of the total for the whole of 2025, with an aggregate value of approximately RMB 250.14 billion, already 182.0% of the whole of 2025.
The value is growing markedly faster than the number of events, which indicates that individual transactions are becoming larger. This imposes more specific demands on transactional legal work: financing structure and equity design, clear definition of intellectual property ownership, due diligence on data compliance, and the allocation of rights and obligations between founders and investors will all be scrutinised more closely as transaction size increases. Periods of transaction activity are often also periods in which disputes and compliance risk accumulate. On the relevant arrangements, see the guide to equity structure for AI start-ups.
Three points for AI companies
Translating the report’s data into action yields three points.
First, calibrate the compliance baseline to the standard of a service for the majority. At this scale, the standards for implementing content labelling, complaint handling, protection of minors and personal information protection should be designed for coverage of the majority of the population rather than for an early, small user base.
Second, bring the new usage scenarios within compliance review. Office work, meetings and image and video processing each carry different data and content risks, and the corresponding handling mechanisms should be built into the product.
Third, advance transactions and compliance together. As the pace of financing accelerates, the regularity of data compliance, intellectual property and equity arrangements will directly affect transaction efficiency and valuation.
The report describes an industry expanding rapidly. For the companies within it, scale brings both opportunity and a matching responsibility. For assistance with filing, data compliance and transaction arrangements, see algorithm filing services.
This article is prepared on the basis of the Generative AI Application Development Report (2026) issued by CNNIC on 29 September 2026 and related reporting by Xinhua and People’s Daily Online. It is for reference only and does not constitute legal advice. Specific compliance assessments should be made in light of the applicable law and the facts.