In mid-2026, statistics from multiple industry media point to a somewhat cruel conclusion: after a year of explosive expansion, more than 90% of companies that entered the AI short drama track have still not achieved profitability, with the vast majority in the stage of "burning money for content, burning money for traffic." At the same time, the daily output of AIGC content is climbing exponentially—a small five-person team, with the help of text-to-video models, can produce hundreds of vertical-screen short drama clips in a single day. Output soaring and profits dropping to zero are happening simultaneously on the same track, which itself is a structural signal worth unpacking.
If we shift our perspective from the "content business" to "software engineering," we will find that the profitability dilemma of AI short dramas is essentially a clearly visible scissors gap in the AIGC content production chain: upstream compute costs are falling but have not yet broken below the break-even line, while downstream distribution and content supply are diluting per-item traffic at an even faster rate. Squeezed from both ends, creators in the middle are left in no man's land.
Text-to-video models are currently the heaviest cost component in the AIGC industry. Taking mainstream video diffusion models as an example, generating a 60-second, 720P short drama clip consumes considerable GPU resources in both the training and inference stages. Over the past 12 months, inference prices from leading vendors have indeed seen cliff-like declines—the cost per generation for some models has dropped by more than 70%, and there has even been a price war with steep discounts on pay-per-use billing.
But behind the price decline there is a detail that is easy to overlook: a short drama is not "a video," but "a video sequence with narrative coherence, character consistency, and continuity of clothing and scenes." Frame-by-frame generation cannot guarantee that the protagonist's appearance remains consistent within 300 seconds before and after. Creators either have to invest a large amount of manpower in post-production face retouching and frame interpolation, or repeatedly generate and repeatedly draw and discard, quietly shifting "generation cost" into "retry cost." Some practitioners estimate that for a finished 3-minute short drama, the total GPU generation time actually invoked is often 20 to 30 times that of the final film. The compute bill has nominally dropped by 70%, but when the actual settlement comes, it does not look as good as imagined.
The conclusion is actually very simple: lower compute prices have lowered the entry threshold for AIGC, but they have not simultaneously lowered the marginal cost of "producing a commercially viable finished product." A low threshold brings more people in, while high marginal costs make it difficult for those entrants to profit. This is precisely the first blade of the scissors gap.
The real impact of AIGC lies in pushing content supply from "handmade" to "industrialized." In the past, a short drama crew, from script and storyboard to shooting and editing, could deliver only a limited number of finished films in a week; now AI teams use "script templates + generation pipelines + batch rendering" to roll out a month's worth of updates for an account in just a few days.
But the problem is that the distribution side has not expanded correspondingly. The total time users spend scrolling short dramas every day is limited, and attention has a hard ceiling. When supply expands at 10x or 20x speed while user time grows only slowly, the traffic, exposure, and payments that each piece of content can receive are naturally diluted. The platform algorithm's dominance further amplifies this squeeze: traffic concentrates toward top hit content, a large number of long-tail AI short dramas sink to the bottom, and conversion rates, completion rates, and ROI decline across the board.
More critically, when content can be copied in batches at almost zero threshold, the scarcity of content collapses accordingly. When scarcity disappears, it means content itself no longer has pricing power, and profit space is compressed to only the thin margins of channels and operations. This is the second blade of the scissors gap: the efficiency dividend on the supply side, in turn, destroys the value anchor of content.
Once this logic is seen clearly, it becomes understandable why leading AI short drama companies have collectively begun to "shift from content companies to technology companies." Simply generating more videos can no longer form a moat, and the real barriers are settling in three directions.
The first is deep customization of vertical workflows. General text-to-video models do not understand the narrative rhythm of short dramas. Teams that understand short dramas use domain data for LoRA fine-tuning, character consistency control, and customized ComfyUI workflows, engineering "generating a usable short drama clip" into a stable and controllable pipeline. This is essentially software development capability, not content creation capability.
The second is data-driven distribution operations. Leading AI content teams no longer bet on a single hit, but build their own data analysis and ad delivery systems, using A/B testing, asset tagging systems, and real-time ROI feedback to decide investment, turning the "content business" into "algorithm-driven traffic arbitrage." What supports this approach is likewise a set of custom-developed management systems and delivery tools.
The third is selling productivity to B-end clients. Since pure content is difficult to monetize, some teams have begun to shift to the toB business of "AI short drama production tools + agency operations," turning the pipeline that cannot make money by supporting themselves into a product that helps others reduce costs and increase efficiency. Content marketing demand from cross-border e-commerce, local life services, and e-commerce platforms is becoming the recipient of this wave of AIGC capabilities.
These three paths share one thing in common: none of them rely any longer on "how much the content itself is worth," but instead on "whether the work of generation, distribution, and operations can be done systematically well with technology." In other words, the focus of competition has migrated from "creativity" to "engineering." Technology itself is not the goal; whether technology can turn a link that originally does not make money into a scalable business is the watershed.
In the short term, this compute-monetization scissors gap is likely to widen further. Because the inference cost of video generation models is still on a downward trajectory, entrants will continue to pour in, the expansion speed of content supply will still run ahead of traffic growth, and the 90% loss structure will be difficult to fundamentally change in the short term.
But signals of mid-term convergence have already appeared. On the one hand, as leading models make breakthroughs in multimodal capabilities, character consistency, and long-video narration, the hidden cost of "retry cost" will be gradually pushed down, and the true marginal cost of compute will eventually fall to a point where high-quality AI content can also generate positive gross margins. On the other hand, platforms are also following up with content governance and premiumization strategies, using algorithms and subsidies to redirect traffic back to high-quality AI content, making "doing it well" once again more worthwhile than "doing more."
It can be predicted that in the next 12 to 18 months, AI short dramas will undergo a round of reshuffling similar to that of short videos and livestream e-commerce in their day—pure generation teams lacking engineering and distribution capabilities will be eliminated first, and those that remain will inevitably be companies that treat AIGC as a software problem rather than a mystical problem to be solved.
For enterprises still watching from the sidelines, this round of "90% not making money" in AI short dramas is not a signal that the industry is a scam, but rather a phased characteristic of a cost structure that has not yet converged. What is truly worth learning from is not betting on an AI hit, but embedding AIGC into existing business chains—for example, using AI to mass-produce marketing assets, product demos, and content seeding, treating it as a cost-reduction tool rather than a money tree. This kind of capability will ultimately settle into code in mini-programs, apps, and enterprise internal management systems, rather than content consumed one time only.
When content is no longer a scarce commodity, using technology to turn generation, distribution, and operations into a sustainably operating system is the truly scarce capability on the content track.
Shenzhen Xiangming Technology Co., Ltd. | Create value with technology | xiangmingit.com
Room 1409, Building C, Shenzhen Century Industrial Center, No. 743 Zhoushi Road, Hezhou Community, Hangcheng Sub-district, Bao'an District, Shenzhen
