Hengdian's Studios Are Empty While Asia's AI Film Pipeline Runs

Hengdian is a small town in eastern China that the film industry nicknamed China's Hollywood. For two decades it hosted hundreds of productions a year and the crews that came with them. This year its sound stages sit idle, because the productions that used to fill them are being generated on servers instead.

The numbers behind that shift arrived fast. Chinese state media reports that more than 95 percent of the roughly 120,000 microdramas released in the first quarter of 2026 were AI-generated, up from close to zero a year earlier. A short AI drama costs a few hundred yuan per minute, against two thousand to twenty thousand yuan for live action, and companies in Hangzhou are using the technology to produce 100-minute features for under 20,000 yuan, about $2,980. The labor market adjusted in the same direction. Chris Shaw, who leads the film school at Lasalle College of the Arts in Singapore, told reporters that a lot of actors and studios have simply disappeared.

Two films, no actors

The clearest proof that this is not a China-only story came from Singapore, which celebrated its 61st birthday this year with two films. One depicts the country under a zombie siege, the other follows a young man who grows up to become Singapore's first astronaut on Mars. Apart from the scripts, everything in both was generated by AI. Neither used a human actor.

The producers behind them make the case in practical terms. Joel Boh, whose studio Edenstone made the four-minute zombie piece, spent about S$8,000, or roughly $6,300, and argued that filmmaking tools have never been this accessible in a century of the medium. Pax Chen, whose studio FizzDragon made the astronaut film, frames it as a diversity fix. Audiences watch films to see themselves, and Asian viewers rarely see themselves reflected in Hollywood productions. AI, in his account, closes some of that gap.

Hong Kong moved in the same direction. In September, broadcaster TVB released a 10-episode short drama built around AI-generated characters, the first of its kind for the network, and has since put its virtual actors into advertising campaigns. Danming Xie, iQiyi's senior vice president and chief AI scientist, put it bluntly: AI is like oxygen, it is everywhere, and the question has shifted from whether to use it to how deeply.

The platform that pivoted

The sharpest signal is what happened to iQiyi. Founded in 2010 as a Chinese answer to Netflix, the company announced in April 2026 that it was pivoting to AI and turning its video app into a social platform hosting generated content. Its founder called AI a once-in-a-decade opportunity. A streaming company deciding that its future is distribution plus generation, rather than acquisition, tells you where the margin is expected to sit.

The catalog reflects the pivot. iQiyi released an AI-generated feature over an hour long in July, the first such film to receive a Chinese online drama distribution permit, and followed it with a series of AI-produced features on the classic Soul Ferry franchise. The first two entries in that series drew more than 8 million yuan in revenue within a month of release, enough to cover both production budgets, and the third arrived on October 1.

Who actually controls the pipeline

A Nomura research note published on October 5 walked through the value chain and found it more concentrated than the volume suggests. The bank's analysts describe ByteDance as dominant across both ends of Chinese AI short drama: its Seedance model sits upstream as the generation tool, and Douyin and Hongguo sit downstream as the distribution channels. Their estimate is that Seedance is used in 80 to 90 percent of production workflows, with one studio putting about 80 percent of its AI video budget into it.

The catch is price. Seedance carries a premium over Kuaishou's Kling and Alibaba's open-source Wan models, and the note argues that studios will switch to cheaper tools as soon as the quality gap narrows. That is a familiar dynamic from every other layer of the AI stack. Whoever has the best model today has it because the alternative is measurably worse, and models close the gap on a schedule of months. Kling, for its part, claims 100 million users and says 70 percent of its revenue comes from outside China, which makes the competition for Chinese studios a two-way street.

Nomura's analysis also shows where the money went as quality expectations rose. With supply exploding, platforms like Hongguo raised their content bar, and the definition of a good AI drama moved from "cheap" to "watchable." In that market, the script or underlying IP now accounts for about 30 percent of total production cost, with compute and production taking more than 20 percent. The tool is no longer the bottleneck. The story is, and the story is the part that still needs people.

The label lands on the platform, not the studio

One thing the industry has to absorb that its predecessors did not is disclosure. China's new rules on labeling generated content took effect on October 1, and they apply to distribution as well as production. Platforms have already updated their terms to require that AI-generated material be marked, and unmarked content faces throttling or removal. For a microdrama pipeline where more than nine in ten titles are machine-made, that means disclosure is no longer optional punctuation on a small share of the catalog. It is the default state of the format.

The practical question is how viewers react. A badge that says a drama was generated is cheap to add and hard to undo if audiences treat it as a quality warning. The platform that raised its content bar is betting that the badge fades into the background once enough good titles carry it. That bet is untested, because the labeling regime is only weeks old and no one has a year of engagement data on labeled versus unlabeled AI drama.

What the industry is actually losing

What the coverage worries about is which jobs survive, far more than whether AI can make a film. The tasks that AI absorbs first are the ones that were already the most repetitive: background performance, crowd work, establishing shots, dubbing. Those are also the entry points where a new actor or technician used to learn the trade. Removing them cuts more than a line item. It cuts the pipeline that produced the next generation of craftspeople.

The counterargument is that the same technology lowers the barrier for people who never got a shot. A filmmaker in Singapore can produce a feature for the price of a used car. A novelist in Chengdu can adapt her own book without pitching a studio. Both of those are real, and neither requires a sound stage.

Whether the audience follows is the test nobody has passed at scale. Microdramas are working because the format is short, fast and consumed on a phone, where production values matter less than pace. A 100-minute AI feature asks viewers to sit with generated faces for an hour and a half, and the bar for that is different. The Singapore films and the iQiyi features are the first real attempts to find out. Hengdian's empty stages are the evidence that the industry has already placed its bet, whether or not the audience agrees.