“The relationship between output and headcount that used to be taken for granted is breaking.”
Start with one number: 44.
That’s the median headcount of companies on the 2026 Entertainment AI 100. The two smallest have 2 and 3 employees respectively.
Within the same application data, one company produces over 100 AI comic-drama episodes monthly with over 3 billion cumulative views; another covers 174 countries and regions with 1.5 million registered users; one makes 93% of its revenue overseas.
The relationship between output and headcount that used to be taken for granted is breaking.
The list was announced live at the Extraordinary Awards · Chengdu AI Entertainment Conference. Here is the complete version.
But this is just the result. What’s worth looking at are the application-data patterns behind it.
The money part is the most counterintuitive, so it comes first.
Of the companies that submitted data, more than half reported no funding or undisclosed funding. At the same time, nearly 90% already have real revenue — 8 companies crossed 100 million RMB in annual revenue, and nearly half exceed 10 million RMB.
This is almost the reverse of the last AI boom. Last time it was: raise money first, find use cases second, talk about revenue last. This time, many companies have reversed the order.
This round of entertainment AI starts not from valuation, but from orders.
The logic is simple. Entertainment is one of the few AI adoption scenarios that generates immediate payment: an AI short drama, an AI song, a tool that produces finished content — sell it and there’s cash flow. No need to wait for the ecosystem to mature, no need to wait for big clients to approve budgets.
But the other side must also be clear. Nearly half of companies have received institutional investment, and top valuations are diverging fast: 2 are above 5 billion RMB, and 4 are in the 1 to 5 billion range. At the same time, some companies are still at “not yet commercialized.” Capital concentration is rapidly moving toward the top — the middle ground will be hard to hold.
By founding year, more than 60% of selected companies were established in 2023 or later, and nearly 40% in 2025 or later. Only 11 were founded in 2020 or earlier; the earliest traces back to 2003.
This structure explains one thing: traditional film and TV companies doing digital transformation are only a small part of this list. The main force is a batch of new companies that, from day one, grew into the shape of AI.
The most convincing example doesn’t need a big company. Wuxian Huimeng was founded in March 2026 with a 2-person team, making AI interactive picture books for kids aged 6–12, where parents and children together generate a custom growth story.
Established companies are also present, and with serious pedigrees: Wondershare, Renderbus, Xiaxu Music, HelloTalk, Xuanjia, and Toga. They’re still in their core business, but the approach has changed. Renderbus grew AI video creation agents out of cloud rendering; Xiaxu Music grew an AI music brand and virtual singer out of game music.
The capacity that the last generation of content companies took 20 years to build, this generation bypasses in two.
That statement holds for both types of company.
Looking at founding teams, you see two types of people converging in the same place.
One type: people who understand models starting to make content. Qingnai Technology is led by Tsinghua and Cornell alumni; Dishi Chuangxiang’s team includes Imperial College and Harvard PhDs and postdocs, plus industry veterans from Huawei and Baidu.
The other type: people who understand content starting to learn models. Lumen Flow’s founding and core team comes from ByteDance and TikTok; Dongfang Qiming’s 11-person core team is assembled from ex-big-tech executives, university professors, and former big-tech core R&D; Yuanqi Gongsheng even turned AIGC applied-talent training into a business, setting up training bases in over a dozen cities nationwide.
Ten years ago, people in entertainment didn’t understand models, and people in models didn’t understand content. Now these two groups have collided — that’s the biggest variable in this year’s list.
By track classification, video generation accounts for 37%, and film/TV/video entertainment another 18% — together over half. Add animation, comics, and music/audio, and content generation and presentation is essentially the main body of this list.
But within the same track, these companies stand in completely different positions.
At the most upstream are the ones building foundational models: Shengshu Technology and Sand.ai are at this layer, competing on controllability of video generation and physical consistency.
The middle layer builds agents “from script to final cut”: Jurilu AI, Enjixiu, Lumen Flow, and MovieFlow. The last one’s self-positioning is blunt — putting the film industry into an ordinary person’s workstation.
Further down are the ones making dramas directly. Xinshijue AIGC produces over 100 AI comic-drama episodes monthly with 3+ billion cumulative views; Enjixiu incubated a 10-million-hit short drama on Hongguo; Guanghe Dongli and HelloShorts ship dramas directly overseas.
There’s also a layer most easily overlooked, doing “the business of drama”: Xingmen Yuandong handles AI short-drama copyright registration, global distribution, and transaction settlement.
Tools solve “making it.” The industry must solve “keeping it going.”
Generating a 30-second video in 2026 is no longer rare. What’s rare is producing a hundred episodes in a row without breaking down, and handling copyright, distribution, and settlement along the way.
Beyond video, this year’s list also surfaced categories that barely existed before: emotional companionship, virtual characters, character intelligence, and real-time interaction engines.
Dongfang Qiming uses a 32-billion-parameter Eastern-culture vertical model, finding adoption in cultural tourism, Eastern wellness, and AI education. Xiaojianbing Technology builds long-term memory and relationship-dynamics systems, aiming to move from “chat responses” to “shared experiences.” Shuban dipal even made hardware — the Dipal D1 is a digital companion that can interact, remember, and grow.
In music too, companies are going deeper. Toga built the world’s first multimodal music generation model, Tianpule, from scratch in 2024, and launched a conversational music creation agent in 2025.
The common thread among this group is that what they sell has shifted from a piece of content to a relationship. That may be the hardest to replace — and hardest to do well — direction in AI entertainment.
A median of 44 means more than half the companies on this list have fewer than 50 total employees.
But their output looks like this: HelloTalk serves 70 million users across 260+ languages; AhaCreator holds a global creator pool of 5 million with 150,000 active accepting creators; OST Media has 5,000+ signed creators across platforms.
At the other end, scale still exists. 11 companies exceed 200 people: Wondershare has 1,575, and both Toga and OST Media have 1,200.
So this isn’t a story of small companies replacing big ones. The real change is that the conversion between “how many people” and “how much content they produce” has been rewritten.
When the marginal cost of content production approaches zero, the unit of a company gets redefined.
Ten people with one agent stack doing the work of an entire department — that judgment is still being debated in other industries, but in entertainment it’s already daily life.
74% of selected companies have overseas or cross-border business.
Target markets, ranked: the US 32%, Southeast Asia 32%, Europe 29%, Japan/Korea 26%, then Middle East 11%, Latin America 11%, Africa 5%.
9 companies make over half their revenue overseas; 5 of those exceed 80%: AhaCreator 100%, Wondershare 93%, HelloTalk 90%, Shuban dipal 80%, Sand.ai 80%.
The last generation of content companies going overseas followed a playbook: prove it locally first, then replicate, pushing into markets with ad-buying budgets and large localization teams. This generation is different — their multilingual adaptation, content structure, and even business models were designed for multiple markets from day one.
HelloShorts is registered in Hong Kong and makes multilingual language-learning short dramas — learn expressions while watching episodes. That product form is hard to serve only one market.
These companies’ default market, from the start, is the whole world.
By city distribution: Hangzhou, Chengdu, Nanjing, Xi’an, Chongqing, Xiamen, and Hong Kong — not just the tier-1 cities.
Tier-1 cities are still the main force, but supply isn’t locked into industry clusters. AI short-drama makers are in Chengdu; AI film/TV creation and distribution platforms are also in Chengdu; AI content creative shops are in Xi’an.
The threshold for this track isn’t entirely about which city you’re in — it’s more about whether you have a group of people around you who truly understand content.
That’s probably why this year’s awards ceremony was placed in Chengdu.
Finally, a word on the evaluation itself.
The 2026 Entertainment AI 100 uses a four-dimensional framework: entertainment innovation and AI-native quality at 30%, experience quality and user engagement at 30%, real market value and growth at 25%, and the last dimension — safety, rights, and ecosystem responsibility — at 15%.
Reserving 15% for rights and safety is because entertainment is one of the industries closest to copyright, minors, and deepfakes. An AI content product can have a high ceiling, but if it doesn’t stand on training-data rights, generation labeling and provenance, and minor protection, it shouldn’t be on this list.
The evaluation also doesn’t look at concept demos. After companies apply, they go through product and content hands-on verification, user and partner research, and then comprehensive review by judges across AI product and technology, gaming, film/TV, music, interactive content, IP and copyright, safety governance, and commercialization.
Publishing the list is just the start. We’ll keep tracking where these companies go next.
If you want to discuss collaboration with these companies, or apply for the next edition, feel free to reach out.
AI entertainment isn’t short of the word “disruption.” What’s worth looking at on this list is what each selected company has actually made of that word.