The finance team got what it asked for. Production costs fell, the loss narrowed, and the quarter looked better than it had in a year. Then somebody read the top line.
Chinese Online (中文在线, SZ: 300364) published its third-quarter report in late October. Revenue for the nine months to September came in at RMB 833 million (≈ US$117M), down 17.64% year on year. Operating cost fell 32.91% to RMB 444 million (≈ US$62.5M). The filing gives one reason: part of its overseas microdrama (短剧) output had been switched to machine production, so the cost of making it dropped.
That sentence is the sector's most useful disclosure this quarter. It is also, read carefully, a warning.
Key takeaways
- Chinese Online's operating cost fell 32.91% after shifting some overseas microdramas to machine production; revenue fell 17.64% over the same nine months.
- Its third-quarter revenue alone dropped 43.95%, a much steeper fall than the nine-month figure.
- Kunlun Tech took the opposite route: revenue up 53.96%, gross margin 74.41%, and a quarterly loss of RMB 1.085 billion.
- Kunlun's sales expense reached 93.09% of its quarterly gross profit. The saved production cost went into buying traffic.
- None of the three largest listed microdrama exporters turned a profit on operations in the period.
The quarter in one table
| Company | Nine-month revenue | Growth | Nine-month result | Sales expense |
|---|---|---|---|---|
| Chinese Online | RMB 833M (≈US$117M) | −17.64% | net loss RMB 132M (≈US$18.6M) | RMB 384M (≈US$54M, −41.84%) |
| Kunlun Tech | RMB 8.937B (≈US$1.26B) | +53.96% | net profit RMB 2.61M; loss of RMB 1.546B (≈US$218M) excluding one-offs | RMB 5.818B (+101.27%) |
| Zhangyue | RMB 2.775B (≈US$391M) | +12.73% | net loss RMB 68.41M (≈US$9.6M) | RMB 673M (≈US$95M) in Q3 |
Three ways to lose money.
What "converted to machine production" actually meant
The switch was not cosmetic. Part of its overseas live-action slate moved to machine production, and sales expense fell 41.84%, which it put down to lower overseas promotion. Gross margin rose to 46.74%.
The result was a smaller business with a better shape. The third quarter on its own was down 43.95%, so the retreat accelerated as the year went on.
This is what a cost programme looks like when demand does not come back. You keep the margin and lose the scale.
The opposite bet, in the same quarter
Kunlun Tech (昆仑万维, SZ: 300418) went the other way and spent.
Nine-month revenue reached RMB 8.937 billion, up 53.96%, with overseas sales at RMB 8.706 billion, more than 97% of the total. Its Tiangong AI (天工AI) business booked RMB 4.321 billion (≈ US$609M), up 168.29%. Third-quarter gross margin hit 74.41%, up 6.35 percentage points.
Then the loss: RMB 1.085 billion (≈ US$153M) in the third quarter, against a profit of RMB 190 million (≈ US$27M) a year earlier. Sales expense for nine months was RMB 5.818 billion (≈ US$820M), more than double. In the third quarter alone it was RMB 2.478 billion (≈ US$349M), equal to 69.26% of revenue and 93.09% of gross profit.
For every hundred yuan of gross profit Kunlun produced in the quarter, it spent ninety-three buying users.
Where the gross profit went
Kunlun's own numbers show the machine working. Its paid platform DramaWave reported that more than 95% of September's new episodes were machine-made. Its free platform FreeReels holds users for more than 80 minutes a day. Sensor Tower put the group's microdrama apps first by monthly active users outside China. September monthly billings on the AI microdrama platforms approached US$100 million, against roughly US$36 million in December 2025.
The content is cheap and the audience is there. But that audience has to be bought episode by episode, at a price set by whoever will overpay that week.
Chinese Online's answer was to stop buying. Kunlun's answer was to buy faster. Zhangyue (掌阅科技, SH: 603533) tried a third way: revenue up 12.73% to RMB 2.775 billion, loss narrowed, with quarterly sales expense held at RMB 480 million, RMB 629 million and RMB 673 million. Restrained, and still a loss of RMB 68.41 million (≈US$9.6M).
The part nobody wants to say out loud
Machine production removed a cost that was never the biggest one. In short-drama economics the dominant line item has long been user acquisition, not shooting. When the cost of an episode falls by two-thirds, the savings are real, but they land in a market where the number of episodes also rises, which pushes the price of attention up.
The industry data agrees. DataEye's 2026 overseas report found locally produced content rising from 32% to 53% of the market in a year, with machine production the main driver, and some platforms approaching 90% machine-made output within nine months. Supply is no longer the constraint. Distribution is.
The three filings describe one transition, not three strategies. The production bottleneck is gone. The acquisition bottleneck has not moved, and now sits on a much larger pile of content.
Honest limitations
All figures here come from third-quarter filings as reported by Chinese financial media; I read the coverage, not the filings, and the companies' segment definitions differ enough that margin comparisons are indicative only. The sales-expense line includes channel fees and brand spending, not just media buying, so "cost of buying users" is a shorthand. I have not seen unit-level retention data, which is what would settle the argument. Companies also choose what to disclose: Chinese Online explained its cost decline in one clause and said nothing about how many titles it shipped.
What readers can do now
- When a content company reports a cost cut, read the revenue line in the same quarter before the nine-month line. The trend inside the period is where the decision was made.
- For any short-drama platform, ask for sales expense as a share of gross profit. Above 70%, the business is renting its audience.
- Treat "95% of new episodes are machine-made" as a supply statistic, not a quality or margin statistic. Ask what happened to cost per paying user in the same period.