China’s AI Short-Drama Flood: Cheap to Make, Hard to Profit

China’s short-drama industry spent 2026 discovering what happens when the cost of making a show falls close to zero. Output exploded, most titles vanished without an audience, and on September 1, 2026, the country’s first dedicated regulation for micro-dramas took effect, including a rule that AI-made dramas carry a notice in every episode.

Volume and profit figures in this story come from different trade bodies, data firms and reporters, and they do not always agree. We name the source for each number rather than blending them into one tidy statistic.

According to estimates from the China Netcasting Services Association (CNSA), reported by CNBC on August 26, 2026, about 128,000 short dramas were released in China in the first quarter of 2026, and more than 95% of them were AI-generated. CNBC described producers ‘flooding the zone’ with titles to see which ones take off.

The second quarter looked different. 36Kr, citing the same CNSA guidelines, reported 239,000 releases in Q2, of which 153,000 were AI dramas. Sixth Tone put the Q2 AI share at 64%, noting that live-action output rebounded. TechNode, adding the two quarters together, counted 367,000 micro-dramas in the first half with AI above 74%, and an average of more than 1,500 new AI dramas a day.

Other counts are narrower. DataEye, as cited by 36Kr, tallied 221,900 new AI and animated dramas on Douyin alone in the first half, more than 1,200 a day. MIT Technology Review, also citing DataEye, reported an average of 470 AI dramas a day back in January. The direction is unambiguous; the exact daily number depends on who is counting and where.

The cost collapse is real. TechNode reported that productions which could run to hundreds of thousands of yuan with traditional methods can now be made for a few thousand, sometimes by a single creator in a few weeks. A FlexTV executive told MIT Technology Review that AI cuts the roughly $200,000 cost of a North American short drama by 80% to 90%.

Making a show and getting it watched are different bills, though. Ashley Dudarenok of ChoZan told CNBC that while production may cost only a few hundred thousand yuan, ‘making the right audience see it can cost far more,’ and estimated that the price of 1,000 promotional ad impressions rose from 50–80 yuan in 2023 to around 150–200 yuan in 2025.

So most titles lose. DataEye’s first-half figures, reported by 36Kr, found that only 0.48% of new Douyin AI and animated dramas passed 100 million views, and taking 50 million views as a break-even line, only 1.3% crossed it. TechNode summed up the period as one where the industry-wide rate of recouping production costs remained low. Treat the break-even line as an industry rule of thumb, not an audited accounting figure.

When thousands of titles chase the same algorithm, they start to look alike. TechNode described platforms flooded with ‘interchangeable AI-generated faces and recycled storylines.’ A screenwriter interviewed by MIT Technology Review described writing ‘the same plot devices over and over again,’ with projects sorted by keywords such as enemies to lovers and rags to riches.

The platforms’ own response tells you how serious the problem became. TechNode reported that ByteDance’s Hongguo is building tools to tune script pacing and cut recycled content at the source, and pre-screening systems that flag repeated AI faces before review.

The Administrative Measures for the Development of Micro-Dramas were approved by the National Radio and Television Administration (NRTA) on July 27, 2026, and published in full by Xinhua, effective September 1, 2026. They cover any series with episodes under 20 minutes, on apps, websites, TV and other channels.

The measures sort dramas into three classes by budget and subject matter, with filing and distribution-permit requirements for the first two and platform pre-review for the lowest class. Article 34 requires that micro-dramas generated or produced with AI carry a notice in a prominent position in every episode, an obligation placed on both producers and broadcasting platforms. Article 37 bars platforms from algorithms that induce addiction or excessive spending.

This is a summary of public reporting and the published text, not legal advice. How strictly the per-episode label is enforced in practice is something the coverage so far does not show.

First, volume is not quality. China ran the experiment of near-free production at national scale, and the result was more shows, not more hits. Cheap tools lower the cost of a bad drama as much as a good one; the scarce resource becomes audience attention, which is getting more expensive.

Second, labeling is coming everywhere. China now requires a notice on each AI episode, and other jurisdictions are moving toward their own disclosure rules; our guide to AI-generated video labeling rules tracks them. Makers who plan disclosure from the start will spend less effort retrofitting it.

Third, story and character are the moat. TechNode’s own conclusion was that the lasting advantage is not the AI tools but the ability to capture audience emotion and control pacing. That matches why the format crossed borders in the first place, which we covered in why vertical dramas travel. For how AI-made and filmed micro-dramas differ for viewers, see AI short drama vs live-action micro-drama.

ShortFling is not in the volume race. Our drama catalog lists 12 vertical series, and none has a playable episode yet; each is marked coming soon, and you can leave a notification to hear when it launches. Every series starts from a written series bible covering the cast, their arcs and the season’s turns, because consistent characters are the part no generator supplies.

What is live today is fiction: serialized stories in separate SFW and NSFW sections, with the NSFW side behind an 18+ confirmation and availability that varies by region and platform. If you would rather write than watch, the AI editor at /write is free within a daily quota, and published work goes through review before it appears.