Last updated: September 28, 2026
In the first half of 2026, 221,900 new AI-made dramas and comic-dramas launched on Douyin, China’s TikTok, according to DataEye. Only 1,055 (0.48%) passed 100 million views, and just 1.3% reached DataEye’s break-even line. AI has made video cheap to produce, but viewer attention has not grown with it. The scarce resources are now distribution, trust, recognizable characters and rights, and the same shift is visible in music, online video and web publishing.
The usual question about AI video is whether it will replace filmmakers. China’s micro-drama industry, the first place where AI video is being produced at industrial scale, suggests a more useful question: what happens to an economy of content when supply grows much faster than the time people have to watch it?
The answer emerging from Douyin is not “AI video fails”. Total viewing is enormous and money is flowing. It is that the rewards are extremely concentrated, prices collapse for everyone else, and the bottleneck moves from making a video to getting anyone to watch it.
What do the Douyin numbers actually show?
The figures come from DataEye’s report on AI dramas for the first half of 2026, first presented at the World Artificial Intelligence Conference in July and cited by Reuters on September 23. Three definitions matter. A “program” is a title, not an episode. The category covers native AI dramas, including photorealistic “live-action” AI dramas and 2D and 3D AI comic-dramas. And the figures cover Douyin only, not every Chinese platform.
In other words, about one title in 77 covered its costs, and 98.7% did not. Yet those titles still drew a combined 515.7 billion plays over the half-year, and DataEye counts more than 500,000 creators. Attention exists in vast quantity. It is just distributed like wealth in a very unequal economy.
The national picture is larger still. Wang Xiaoliang of China’s National Radio and Television Administration said on September 17 that 430,000 micro-dramas went online across China in the first eight months of 2026, 13 times as many as in all of 2025, and that more than 90% were AI dramas. China has over 800 million micro-drama viewers, and the regulator put the 2025 market above 100 billion yuan. By every estimate we found, micro-dramas now earn more than China’s box office, which was 51.8 billion yuan in 2025.
Why is cheap production not the same as profitable production?
Because when everyone’s costs fall, prices fall faster. Reuters, citing state broadcaster CCTV, reports that the per-minute price for AI short-drama production dropped from about 5,000 yuan to “just a few hundred yuan” in the first half of 2026. That figure is the price outsourced studios quote for standardized work, which makes it a sign of oversupply as much as efficiency. DataEye puts the cost of higher-quality AI drama at roughly 800 to 1,500 yuan per minute, still about a fifth to a seventh of live action.
| Indicator | Before | Now | What it means |
|---|---|---|---|
| Standard AI drama production price per minute (CCTV) | ~5,000 yuan | “a few hundred” yuan | Price war among suppliers |
| Revenue per 10,000 plays (Yicai, citing industry data) | 30–100 yuan (H2 2025) | 5–10 yuan (2026) | Each view earns roughly 90% less |
| Compute cost per AI comic-drama episode (Yicai) | 17 yuan | 80–100 yuan | Higher quality demands more generation |
| Active contracted producers (Yicai) | 1,216 | 698 | 42% of studios dropped out |
| Wedding scene, one studio’s example (Reuters) | 60,000 yuan conventionally | 1,400 yuan with AI | The cost gap that started the rush |
Read together, these rows describe a margin squeeze. Revenue per view is falling, the compute bill per episode is rising as studios compete on quality, and the price buyers pay for production is collapsing. One Chinese business outlet estimates that buying traffic accounts for around 70% of industry costs; that figure comes from a single source, but it is consistent with everything else here. If distribution is the expensive part, making the video cheaper barely moves the total.
This is classic economics, not a new AI phenomenon. When a good becomes cheap and easy to produce, supply expands until the price falls toward the marginal cost of the cheapest producers. We explain the mechanism in our guides to market equilibrium and price elasticity of demand. The difference with content is that demand is capped by human time, which does not grow when supply does.
Why are the winners so few?
Attention markets follow power laws. A 2026 arXiv study of creator earnings on Patreon, covering more than 100,000 creators, found a distribution with a power-law exponent of about 2, “an extremely heavy-tailed distribution and a strong ‘rich-get-richer’ dynamic”; YouTube-linked creators were the most concentrated. The researchers also found that stronger power laws go with weaker creator middle classes. The DataEye funnel is a clean real-world example: most titles get little, a small tail gets almost everything.
Abundant supply intensifies this. Recommendation algorithms have more candidates to test, the average title gets less exposure, and early signals of engagement decide which few get amplified. Some Chinese producers now describe their strategy as releasing many cheap titles and letting the audience pick, which makes the platform’s algorithm, not the studio, the real editor.
Why are Chinese cities subsidizing a glutted market?
Because local governments see AI video as an industry to win, and industrial policy tends to reward capacity. Reuters reports that Shenzhen offers studios subsidised rent and technical support; Shanghai introduced measures in May including compute support, a pilot production base and backing for AI micro-drama “one-person companies”, with support of up to 10 million yuan per project; Beijing’s Huairou district offers compute vouchers; and Hainan offers subsidies and rent waivers. Hangzhou’s districts have added a 500-million-yuan fund in Shangcheng and an “AI Film Valley” in Xihu, launched on September 16.
Reuters notes signs that AI video “may already be approaching overcapacity”, drawing the familiar comparison with other subsidized Chinese industries. Subsidies lower the cost of entry, which increases the number of studios competing for the same fixed pool of attention. For any individual studio, that makes the economics worse, not better.
What are the rules on AI video in China?
They are tightening in three ways at once:
- Labelling. China’s measures for labelling AI-generated content took effect on September 1, 2025, requiring visible labels and embedded metadata. A new micro-drama regulation from the NRTA took effect on September 1, 2026; its Article 34 requires AI-made micro-dramas to carry a prominent label in every episode.
- Enforcement. The NRTA says 68,000 non-compliant micro-dramas have been removed so far in 2026, more than 90% of them AI dramas, and more than 1,200 accounts handled.
- Likeness. Chinese actors have raised concerns about AI use of their faces and voices, and an actors’ committee linked to the industry federation has warned against unauthorized use. Reports describe a Supreme People’s Court guidance in September treating a recognizable likeness as infringement. Reuters notes that China still lacks clear copyright rules for AI-generated content.
In the EU, the comparable obligation is the AI Act’s Article 50 on transparency for deepfakes and AI-generated content. China’s per-episode rule is more prescriptive. Our EU AI Act guide explains what European publishers and platforms need to disclose.
Is the same thing happening outside China?
Yes, in different forms, and the pattern is consistent: generation gets cheap, platforms respond by filtering, and attention concentrates.
| Market | What happened | Lesson |
|---|---|---|
| OpenAI’s Sora app | Launched October 2025; downloads peaked at about 3.3 million in November 2025 and fell to about 1.1 million by February 2026 (Appfigures). The app closed on April 26, 2026, and the API was discontinued on September 24. | Easy generation did not create lasting audience demand for a standalone AI-video feed |
| Spotify | Removed more than 75 million “spammy” tracks over twelve months (announced September 2025) and introduced AI disclosure and impersonation policies | Platforms filter mass-produced supply rather than reward it |
| YouTube | In July 2025 clarified that its monetization policy excludes content that is “repetitive or mass-produced” | Monetization follows originality, not volume |
| Hollywood | SAG-AFTRA’s 2026 contract, ratified with 91.42% support, tightens consent and compensation rules for digital replicas; studios’ copyright suit against Midjourney continues | Likeness and training rights become priced assets |
| Web publishing | Graphite estimates that about half of new English-language web articles are now mostly AI-written, but that these articles “largely do not appear in Google and ChatGPT” | Supply doubled; visibility did not |
Google’s search changes point the same way. Its March 2024 policy against scaled content abuse targets content produced at scale to manipulate rankings “whether automation, humans or a combination are involved”, and Google said the update led to 45% less low-quality, unoriginal content in results. The target is not AI; it is mass production without added value.
What does this mean for publishers like us?
DecodeTheFuture is a publisher, so this is not an abstract question. If the cost of producing an article or a video approaches zero, the value of any single piece of generic content approaches zero too. What keeps value when production is abundant is what cannot be produced at scale:
- Selection. Deciding what not to publish, and saying clearly what is confirmed and what is not. A verified fact table is harder to mass-produce than a fluent summary.
- Original data and analysis. Numbers you calculated, tested or collected yourself, like the DataEye funnel reframed as a break-even rate, or a cost model a reader can run.
- Trust and track record. Readers and search systems increasingly weigh who stands behind content, which is why authorship, sourcing and corrections matter more, not less.
- A direct audience. Newsletters, returning readers and communities are distribution you do not have to buy from an algorithm.
- Recognizable voice and characters. The Douyin titles that break through tend to have consistent characters and formats. For publishers, that means a consistent editorial point of view.
The lesson from 221,900 AI shows is not that AI content is worthless. It is that AI turns content from a scarce product into a commodity, and commodities compete on cost until the only durable advantage is something the competition cannot copy.
How should studios and creators measure success now?
View counts become a weak signal when supply explodes, because the average falls even when the total rises. Better measures are the break-even rate (the share of titles that recover their cost, 1.3% in DataEye’s data), revenue per thousand views after traffic-buying costs, audience retention across episodes, and the share of viewers who return for a second title. For platforms, the metric that matters is whether more supply increases total viewing time or merely splits the same time across more titles. China’s numbers suggest both are happening at once, and the balance will decide whether AI video becomes a durable industry or a subsidized glut.
FAQ
How many AI dramas were released on Douyin in 2026?
DataEye counted 221,900 new native AI drama and AI comic-drama titles on Douyin in the first half of 2026. Only 1,055, or 0.48%, passed 100 million views.
What is an AI micro-drama?
It is a short serialized drama, with episodes of a few minutes, made largely with generative AI for video, images, voices and scripts. China’s new micro-drama rule defines micro-dramas as series with episodes shorter than 20 minutes.
Are AI micro-dramas profitable?
For a small minority. DataEye uses 50 million plays as a break-even line, and only 2,886 of 221,900 titles, about 1.3%, crossed it in the first half of 2026. Revenue per 10,000 plays has also fallen by around 90% since late 2025.
How much does it cost to make an AI drama in China?
CCTV reported standardized per-minute prices falling from about 5,000 yuan to a few hundred yuan in 2026. Higher-quality AI drama costs roughly 800 to 1,500 yuan per minute, about one fifth to one seventh of live action.
Does China require AI videos to be labelled?
Yes. AI-generated content labelling rules took effect on September 1, 2025, and from September 1, 2026 AI-made micro-dramas must carry a prominent AI label in every episode.
What happened to OpenAI’s Sora app?
OpenAI closed the Sora app on April 26, 2026, after downloads fell sharply from their late-2025 peak, and discontinued the Sora API on September 24, 2026.
Sources and further reading
- Reuters (via Investing.com), China fuels rush to turn AI video into an industry.
- The Paper, DataEye H1 2026 AI drama report (Chinese).
- TVOAO, DataEye AI drama and comic-drama data (Chinese).
- Yicai via Sina, AI drama revenue and producer data (Chinese).
- Jiemian, CCTV report on AI drama production prices (Chinese).
- China News Service, NRTA press conference on micro-dramas (Chinese).
- Cyberspace Administration of China, Measures for labelling AI-generated content (Chinese).
- China Film Administration, 2025 box office (Chinese).
- Shanghai Municipal Administration of Culture and Tourism, AI micro-drama measures (Chinese).
- The World of Chinese, Actors selling their likeness to AI micro-dramas.
- Strauss, Yang and Mazzucato, Rich-get-richer? Platform attention and earnings inequality (arXiv 2509.26523).
- TechCrunch, OpenAI’s Sora app is shutting down.
- Music Business Worldwide, Spotify deleted 75 million spammy tracks.
- YouTube Help, Channel monetization policies.
- Google, March 2024 core update and spam policies.
- Graphite, AI now writes as many online articles as humans do.
DataEye figures refer to Douyin only and to titles, not episodes. Yuan amounts are reported values; Chinese sources were checked for unit conversions (1 yi = 100 million).

