Why China’s AI Short-Drama Bubble Burst: 220,000 Shows, 98.7% Never Earned Their Money Back

In the first half of 2026, China produced roughly 220,000 AI-generated short dramas. By the industry’s own accounting, 98.7% of them failed to recoup their production costs. A market that was supposed to be the next gold rush turned into a volume game where almost everyone loses. The collapse was not caused by one failure. It was caused by four, stacking on top of each other.

The Japanese-language analysis that first surfaced this figure points to a structural trap: AI made production nearly free, but distribution stayed expensive. The result was a supply shock that destroyed the economics of the entire category.

1. Production Became Almost Free. Attention Did Not.

The starting point was genuine cost collapse. AI video models cut short-drama production costs by 80% to 90%, bringing them down to 2% to 10% of traditional live-action budgets[reference:1]. A mid-tier AI short drama could be made for around 20,000 yuan, compared with 200,000 yuan for a live-action equivalent. Some animated “manju” (漫剧) productions were made for a few thousand yuan.

Cheap production triggered an entry stampede. In the first half of 2026 alone, Douyin added 221,900 AI dramas and manju titles—roughly 1,200 new shows per day, or one every 70 seconds. By the first eight months of the year, China had produced 430,000 micro-dramas, 13 times the previous year’s total, with about 90% AI-generated

But attention did not scale with supply. Of the 221,900 titles on Douyin in H1 2026, only 1,055 crossed 100 million views—a hit rate of 0.47%. Only 1.3% reached the industry’s 50-million-view break-even line[reference:6]. The market grew to an estimated 400 billion yuan in 2026, up 138% year on year, but over 90% of participants had already exited by September

As one producer put it: live-action brought in maybe 1,000 companies; AI lowered the barrier and 10,000 showed up. “Costs dropped by an order of magnitude, and so did the number of competitors—up by an order of magnitude”.

2. The Money AI Saved Went Straight to Traffic Bidding

This is the part that killed the economics. AI eliminated production costs, but it did not eliminate the need to buy distribution. On China’s free-ad-supported short-drama platforms, traffic bidding (投流) consumes 80% to 90% of total cost, while compute accounts for about 10%. What is left for the creator is roughly 5%.

The revenue side moved in the opposite direction. “Per-ten-thousand-views” payout (万播分账)—the core revenue metric for free-ad platforms—fell from 30–100 yuan in 2025 to 5–10 yuan by November 2026. Tail-end, assembly-line productions earned as little as 1–2 yuan per 10,000 views

The core contradiction: AI reduced production cost by 80–90%, but traffic cost rose by more than 100% year on year, and revenue per view collapsed by 60–70%. Every yuan saved on production was reclaimed by the auction for attention.

3. The Platform Subsidy That Built the Bubble Was Pulled Away

The AI short-drama boom was never purely market-driven. It was subsidized, and the subsidy was withdrawn.

Hongguo Short Drama (红果短剧), the ByteDance-owned free short-drama app that dominates the market with over 300 million MAU, introduced a model that changed the industry: free viewing with 3–7 second ads between episodes. It captured the “willing to spend time but not money” demographic and pushed average daily time-on-app above 120 minutes.

Then the guarantees disappeared. Hongguo cancelled its guaranteed payment (保底分账) for AI short dramas, keeping only a 20% base revenue share plus performance-based incentives. Douyin and Kuaishou followed with tighter policies[reference:14]. In August 2026, Hongguo cut AI script guarantees by up to 75%, eliminating minimum payments for A-tier and A+ scripts entirely.

Once the floor was removed, the true economics became visible. Under a pure revenue-share model, a show that does not hit scale earns almost nothing. And most shows do not hit scale.

4. The Content Was the Same, and Users Noticed

AI solved production, not storytelling. The result was mass-produced homogeneity: the same “comeback,” “mother-in-law,” and “wealthy family” plots, re-shot with different titles and different AI-generated faces that looked eerily similar across shows.

The audience response was brutal and measurable. Peak completion rates on Hongguo’s AI manju section halved from 42% to 19%. Average daily viewing time fell 23% quarter on quarter. AI micro-dramas’ completion rate sat below 15%, compared with over 40% for top live-action short dramas. One blogger watched her four new AI shows drop from 42% to 19% completion—none recouped costs.

The supply-side shakeout followed. Active AI manju production companies fell from 1,216 to 698 in the first quarter of 2026 alone[reference:20]. More than 5,200 low-quality titles were taken down in six months.

Regulation added another layer. From July 2026, AI micro-dramas were placed under a three-tier filing system based on investment size. From September, AI-generated episodes must carry visible AI labels, and platforms are banned from using algorithms that induce addiction. For a category built on low-cost volume and algorithmic push, compliance costs rose while the growth mechanics were restricted.

What Survives: The 1.3% and the Export Pivot

The bubble did not kill the category. It killed the volume strategy.

A small number of titles proved the model can work when distribution is earned rather than bought. 持械入宋, a zero-budget AI drama made by a 54-year-old first-time creator, crossed 12 billion views across platforms without paid traffic and reportedly earned over 10 million yuan. Its success was not replicable at scale—it depended on IP selection, narrative pacing, and organic distribution that most producers cannot manufacture.

The structural escape route is overseas. By September 2026, AI titles accounted for over 60% of China’s exported micro-dramas. The overseas micro-drama market is projected to exceed $6 billion in 2026, with Chinese companies holding nine of the top ten paid short-drama apps globally. But export is not a free lunch: overseas user acquisition costs in North America reach $5.28 per user, and traffic buying still accounts for about 50% of total overseas cost.

The shift is from “selling shows” to “exporting the industrial chain”—platforms, tooling, and production capacity. That is a more durable business than betting on the next AI-generated hit.

The Lesson

The AI short-drama collapse is a case study in a specific mistake: mistaking a production-cost breakthrough for a business-model breakthrough. AI removed the moat around filming. It did not remove the moat around distribution, attention, and storytelling. In a market where supply can double every quarter and attention cannot, the savings from technology flow to whoever controls the bottleneck. In this case, that was the traffic auction and the platform.

The 98.7% failure rate is not a story about AI being bad at making videos. It is a story about what happens when everyone can make the product and no one can make the audience.