China Tops 700 Million Generative AI Users as Office Tools, Computing and Investment Expand

More than half the population now uses generative AI

China had more than 700 million generative artificial intelligence users by June 2026, with adoption exceeding 50% of the population, according to the China Internet Network Information Center. The expansion has brought AI into everyday questions, office work and creative tasks while increasing demand for computing infrastructure and investment.

The center's Policy and International Cooperation Institute released its Report on the Development of Generative Artificial Intelligence Applications (2026) on September 29 at the seventh China Internet Infrastructure Resources Conference. An official government summary gives a picture of growth extending from consumer applications to factories, agriculture and robots.

China ended 2025 with 602 million generative AI users after adding 353 million during that year. About another 100 million joined in the first half of 2026. Adoption rose from 17.6% in 2024 to 42.8% in 2025 before passing the halfway mark in 2026.

Generative AI creates material such as text, images, music, video and computer code in response to instructions. The report describes a further shift: developers are building systems that can carry out connected tasks and deliver finished work, rather than simply produce an answer.

The figures also show how unevenly that shift is developing. Answering questions remains much more common than preparing presentations or meeting minutes. Meanwhile, financing has grown faster than the number of investment transactions, and robotics companies account for a rising share of the business base.

How quickly has the user base grown?

The increase of 353 million users during 2025 implies a base of about 249 million at the end of 2024. On those figures, China's generative AI audience more than doubled in a single year.

The main milestones show how quickly adoption moved from a minority activity toward mass use:

• End of 2024: about 249 million users, with adoption at 17.6%.
• End of 2025: 602 million users, with adoption at 42.8%.
• First half of 2026: more than 700 million users, with adoption above 50%.
• September 29, 2026: the center released its annual application development report.

Young people remain a large part of the audience. Users aged 19 and under accounted for 26.4% of the total at the end of 2025, the largest age group identified in the report.

Older users are becoming more prominent too. Figures published in February put the share aged 40 and above at 30.3%, up 4.9 percentage points in six months. Those age categories cover different spans, but both figures help explain why developers are targeting uses beyond student questions and youthful experimentation.

Questions still dominate, while office use accelerates

Some 76% of generative AI users said they used it to answer questions. Processing images or video followed at 47.8%, while 37.6% used it for text tasks and 32.5% for work summaries, meeting minutes or presentation slides.

Question answering was therefore more than twice as common as the grouped office tasks. Users could report several activities, so the percentages describe overlapping uses rather than separate groups that add up to 100%.

General AI assistants and office productivity applications each recorded growth of more than 100% in usage compared with a year earlier. That measure concerns how often applications were used, rather than the number of people using them.

For an office worker, the practical change is the move from asking how to write a summary to asking a system to produce one from working material. For developers, repeated use creates a business opportunity: tools that become part of a daily routine can generate more demand than occasional experiments.

From conversation to completed assignments

Coding became a leading measure of model performance worldwide from the second half of 2025, according to the report, and major companies made it a central research priority. By the first half of 2026, leading Chinese models could deliver complete software engineering projects, the report says.

AI agents are central to that transition. An agent is a system designed to pursue a task through several steps, potentially using software tools along the way. A conventional chatbot might explain how to make a presentation; an office agent aims to assemble the presentation itself.

Office agents are the most prominent form of agent deployment for ordinary users, the report says. Improvements in coding, office tasks, image creation and video generation have also increased willingness to pay.

Developers earn revenue when customers call their models, meaning they send requests for the systems to process. Those requests can come through consumer applications or software built by other businesses. The report identifies model usage as an important source of revenue growth for both large companies and startups.

The financial scale of those gains remains unspecified: the report describes rising usage and payment demand without giving a total revenue figure for model calls.

Computing capacity expands behind the applications

China's intelligent computing capacity reached 2,185 EFLOPS by June 2026, up 177% from a year earlier. That means capacity was about 2.77 times its level a year before.

EFLOPS measures computing capacity in units of one quintillion floating point operations per second. Floating point operations are calculations involving numbers with decimal points, a basic part of training and running AI models.

The report also records the opening of China's first fully domestic AI supercluster with 100,000 accelerator cards. These cards are specialized processors that perform large numbers of calculations together. Connecting them into a cluster allows demanding AI workloads to be spread across many processors.

The cluster uses domestic technology across chips, computing, storage, networking and cooling, according to the report. The breadth matters because operating a large AI system requires more than processors: data must move quickly, results must be stored and equipment must remain cool.

China had also built more than 120,000 high quality datasets by the middle of 2026. Datasets are organized collections of material used to develop or assess AI. Their expansion accompanies the growth in computing and models.

Domestic models reach deeper into industry

DeepSeek and Moonshot AI released multiple open source models with parameter counts in the trillions, according to the report. Parameters are numerical settings learned during training that help a model produce responses.

The report says Chinese teams developed all six leading models on a major global model usage ranking, although the official summary does not identify that ranking. It also describes closer adaptation between domestic models and Chinese computing chips.

Industrial applications now cover steel, nonferrous metals, electricity and telecommunications. Companies are using AI in product development, quality inspection and customer service, extending its role beyond consumer chat applications.

More than 30% of manufacturing enterprises above the designated size threshold were using AI technology by June. That statistical category refers to businesses meeting an official size requirement, rather than every factory. Applications stretched across research, design, production, inspection and equipment maintenance.

China had also developed 16 national advanced manufacturing clusters focused on digital industries, including integrated circuits and AI. Together, these figures show how the consumer adoption story is accompanied by organized deployment in production.

Robotics gains ground among AI businesses

Corporate registration information collected through June identified 10,740 Chinese companies working in large models, AI agents, robotics and related fields. Guangdong, Beijing, Shanghai, Jiangsu and Zhejiang accounted for 76.5% of them, concentrating more than three quarters of the business base in five locations.

Companies established within the previous three years represented 18.8% of the total, up 9.8 percentage points from a year earlier. Businesses aged five to ten years formed the largest group at 41.4%, while those at least ten years old accounted for 29.7%.

Robotics and embodied intelligence companies made up 32.9%, an increase of 11.2 percentage points. Their previous share was therefore 21.7%, showing a substantial change in the composition of the sector.

Embodied intelligence puts AI into physical machines that sense and act in their surroundings. The report describes applications in industrial production, surgery and everyday services, where software must coordinate movement as well as process information.

China had more than 400 humanoid robot products in the first half of 2026. The report forecast full year production above 100,000 units. The product figure counts different products, while the production forecast concerns the number of machines expected to be made.

Investment rises faster than transaction numbers

China recorded 1,255 AI investment and financing transactions in the first half of 2026, with a combined value of about 250.14 billion yuan. The number of transactions equaled 78% of the total for all of 2025, while their value reached 182% of the previous full year's amount.

In six months, financing value was already 82% above the amount recorded during the whole of 2025. Using the report's rounded percentages, average value per transaction was about 2.3 times the 2025 average. More money was therefore being committed per recorded transaction, alongside the expanding flow of deals.

Consumers were also spending on devices. Some 38.7% of Chinese internet users bought smart hardware online during the preceding six months. Wearables were purchased by 20.2%, while 18.2% bought consumer electronics such as smartphones or tablets.

Those categories can overlap because a person may buy both. The report identifies wearables and consumer electronics as the main entry points into smart hardware, linking the expansion of AI applications with devices people carry or wear.

Agriculture and regulation move toward practical use

The State Council's June 2026 plan to accelerate agricultural and rural modernization during the 15th Five Year Plan period made AI applications and smart agriculture an important objective. Agricultural models released during the preceding year included Shennong 4.0, Guyu Agricultural Meteorology V2.0 and Fanli 4.0.

Agriculture joins manufacturing, office work and scientific research among the fields highlighted in the report. The policy focus is moving from solving technical problems toward deploying systems for particular tasks while developing rules for their use.

The State Council's guidance on implementing the AI Plus initiative provides the broad policy framework, with supporting measures covering industrial applications, infrastructure and international cooperation.

By June, 988 generative AI services had completed filing with the Cyberspace Administration of China. That service count measures a different part of the market from the 10,740 companies identified through registration information: companies and filed services are separate categories.

The expansion of agents also broadens the safety agenda. The report warns that risks now extend beyond generated content and information distribution to cybersecurity, protection of critical infrastructure and control over the most capable models.

Scientific capabilities bring a sharper debate

The report recounts a September announcement by OpenAI that an AI agent system had solved a major mathematical problem in 88 hours. It also describes a response from 25 Fields Medalists in a joint statement titled The Serious Misalignment of Artificial Intelligence in Mathematics.

According to the report's account, the mathematicians acknowledged substantial recent improvements in mathematical ability and the capacity to solve major open problems, while arguing that the approach was harmful to mathematics and its community.

The episode connects the report's central themes: systems are being assessed by the work they complete, and growing capability brings disputes about how that work should be conducted. In offices, laboratories and physical infrastructure, the decisions increasingly concern both what AI can do and who controls its actions.

Key Points

• China exceeded 700 million generative AI users in the first half of 2026, with adoption above 50%.
• Question answering remained the leading use at 76%; assistant and office application usage each more than doubled.
• AI computing capacity reached 2,185 EFLOPS, up 177% from a year earlier.
• AI investment and financing reached about 250.14 billion yuan across 1,255 transactions in six months.
• Robotics and embodied intelligence represented 32.9% of the identified AI company base.
• The report describes a shift toward completed tasks alongside broader concerns about cybersecurity, infrastructure and model control.