China's AI video boom: 221,900 shows, 1,055 hits
Chinese cities are paying studios to make AI video at industrial scale, and the audience has not kept up. Also: a Chinese startup opens a crowdsourced robot-training market, and the FT prices the neolab wave.
Chinese local governments are competing to subsidise AI video production, and the output has already outrun the audience: 221,900 new AI shows launched on Douyin in the first half of 2026, and only 1,055 of them passed 100 million views, the benchmark the industry uses for a hit. That ratio comes from DataEye and was reported by Reuters on Friday, alongside the cost curve that explains the flood — a minute of AI short drama fell from 5,000 yuan (about $747) to a few hundred yuan in six months, per state broadcaster CCTV. A wedding scene that costs 60,000 yuan to shoot conventionally costs 1,400 yuan in AI, film student Pan Xiaojun told Reuters. When the marginal cost of a show collapses by more than 90%, the number of shows stops measuring demand and starts measuring supply — and here the supply is being paid for by the state.
The subsidies are itemised. Beijing's radio and television bureau published a measure on September 24 covering 2026–2029 that pays up to 3 million yuan per project for AI audiovisual technology, content and scenario work, and 1 million yuan for safety and services, with content grants judged on theme, production level, duration and viewership. The grants are retrospective: they reward projects that already produced results rather than funding a pitch, which turns the programme into a scoreboard for what the city considers a success.
Below the municipal level the incentives get more specific. Huairou, Beijing's film-industry hub, offers micro-drama producers compute vouchers worth 30% of computing costs, capped at 20,000 yuan per title and 26,000 yuan if the finished work is distributed overseas. Shenzhen funds AIGC micro-dramas at up to 30% of project cost and 2 million yuan for a first producer, and gives AI film talent subsidised rent. Shanghai and Henan run their own schemes, and Beijing holds a 260 million yuan audiovisual technology fund. iQIYI's chief executive says the platform is "all-in" on AI and pays some creators a subsidy for AI content that streams on it.
The warning signs are the ones China's EV and solar buildouts taught everyone to look for. Chinese industry data puts roughly 90% of AI short-drama companies in the red, and the sector's breakout rate of 0.47% is capacity arriving years before demand. The National Film Administration has granted a public screening licence to "Sanxingdui: Future Memories," a 90-minute science-fiction feature from Bona Film Group and the first AI film from a major Chinese studio cleared for theatrical release — a real milestone, and one that arrives while actors object to the use of their likenesses, voice actors lose work, and audiences complain about plagiarism. China requires labels on AI-generated content but still has no clear copyright rule for it, which is the gap European regulators are watching rather than filling. We covered the earlier milestones in China's prime-time TV slot now belongs to an all-AI drama.
A Shanghai startup wants to turn robot training data into gig work, and says 20,000 people signed up before launch. Mifeng Technology unveiled its crowdsourcing platform on September 23: download the app, claim a collection device, and get paid by verified hour for recording ordinary tasks — making milk tea, folding clothes, caring for pets — in your own workplace. The company says it has produced 20,000 capture units — the industry's first mass production run of such devices — and accumulated 1 million hours of "bodyless" data across 22 scenario categories, already sold to Tencent's Robotics X lab and Ant's Lingbo team. In a one-month beta, 20,000 users filed 13,000 collection tasks, and one participant earned more than 5,000 yuan in a month.
That is Figure's Index model with one structural difference: Index feeds Figure's own models, while Mifeng sells processed data to whoever buys. It is also the data layer of Chinese embodied AI becoming an industry of its own — a shift we first covered in August, when China Telecom led funding for embodied-AI data startup Mifeng. The open question is quality: the platform's answer is to grade rather than filter collected data, which is a reasonable engineering compromise and also an admission that most of what a crowdsourced network captures is not directly trainable.
AI neolabs — the elite-team startups, many without products, markets or revenue — raised $24 billion in the past two quarters, nearly five times what OpenAI and Anthropic raised in the years up to ChatGPT, the Financial Times reported Friday, citing data from Radical Ventures. The FT's own Lex column flags the caveat that "these numbers can hide some sleight of hand," and Radical's June report publishes the ratio rather than the dollar figure: it puts OpenAI and Anthropic at $5.3 billion combined between OpenAI's founding and ChatGPT, and says the neolab wave raised roughly seven times that in the last year and nearly five times that in the last two quarters alone. The FT names Safe Superintelligence, which reached a $32 billion valuation last year without shipping anything resembling a product, and Reflection AI, which went from $105 million at a $555 million valuation to 25 times as much at $25 billion in about a year. We have covered the same pattern from the inside — Jeff Dean's Discovery Loop seeks a $50B valuation weeks after its $1B raise. The read worth keeping: at this stage of the cycle the money is not pricing customers, it is pricing access to compute and to a handful of researchers.
What to watch: whether Beijing publishes the first round of AI content awards — retrospective grants reveal what "good results" means in practice — and whether Figure and Mifeng ever publish comparable data-quality numbers, which is the only way to know if paid collection produces training-grade data or just volume.
China is treating AI video the way it treated solar panels: subsidise capacity, accept a brutal shakeout, and win on cost. Is that industrial policy working as designed, or a bubble with a budget line? Tell us in the comments.
Sources: Reuters · Beijing Municipal Government — AI+audiovisual project support measure, 京广发〔2026〕71号 · Huairou District — compute vouchers for AI micro-dramas · Shenzhen Municipal Government — micro-drama support rules · TMTPost · QbitAI · Shanghai Securities News · IT Times via Guancha · Financial Times · Radical Ventures — The Rise of the NeoLab · Techmeme
