It has been a while since OpenAI shut down one of my favorite apps , Sora, apparently because the economics were simply unsustainable. Sora reportedly cost an estimated $15 million a day to run and generated just $2.1 million in lifetime revenue.

Into that vacuum stepped five Chinese integrated video AI stacks that are more capable, cheaper and more commercial than anything OpenAI has shipped: ByteDance’s Seedance , Alibaba’s Wan and Happy Horse , Kuaishou’s Kling , MiniMax’s Hailuo AI and Tencent’s Hunyuan. Together, they are powering an industrial-scale AI-generated content economy that now produces 470 AI-made micro-dramas every day.

How these Chinese labs won the video AI race is not just a story of better engineering, though the engineering is genuinely remarkable. It is a story of structural advantages: data scale, vertical integration, government support, aggressive pricing and a domestic market that has become the world’s most demanding audience.

Here are the five forces that explain why, in one of the most commercially consequential AI capabilities of 2026, Silicon Valley is no longer leading.

1. The Data Moat: Training on Everything, Constrained by Nothing

The single most important competitive advantage these Chinese video AI labs have is training data. The volume, diversity and labeling quality of the video footage used to train their models is enormous, and they operate under a very different legal regime from their U.S. counterparts.

In the U.S., AI labs face constant intellectual-property litigation. Getty Images sued Stability AI. The New York Times sued OpenAI. Disney and Universal sued Midjourney. The practical result is that many American labs have become far more risk-averse about the data they use for training.

By contrast, ByteDance’s Seedance can draw on the vast video corpus of Douyin, the Chinese version of TikTok, which generates billions of clips each month. Kling benefits from Kuaishou, another massive short-video platform. The same theory applies to YouTube for Google.

2. Vertical Integration: The Lab IS the Platform IS the Studio

Runway ML is an excellent video AI company, but at its core it is still a tools company built for creators and Hollywood studios.

ByteDance, Kuaishou, and Tencent are not just tools companies. They are vertically integrated entertainment platforms that also happen to build models. ByteDance built Seedance to power Douyin’s creator economy and its Red Fruit micro-drama platform. Kuaishou built Kling to support its own short-drama studios. The content generated by these models is deployed immediately across their own apps at scale.

That also means the business model is fundamentally different. It is not centered on subscription fees from individual creators. It is driven by advertising, paid drama subscriptions and virtual goods sold to billions of users inside their own apps. It is unimaginable for any Western AI Labs to treat Inference cost as marketing expense, one that drives content creation on their own app, which drives advertising revenue.

That video AI flywheel simply does not exist in the West, and it would be very difficult to replicate.

3. Micro Drama Is The First Industrial Scale Use of Video AI

Micro-dramas are the first true mass-market use case for AI-generated video, with hundreds of millions of Chinese users watching and a new AI short drama created roughly every 90 seconds. And yes 90 seconds is not a typo.

A micro-drama is a serialized, short-form video designed for vertical mobile viewing, typically 60 to 90 seconds per episode and 80 to 100 episodes per series. The genre emerged around 2020 and by 2023 was already generating more revenue in China than the domestic theatrical box office. By 2025, Deadline reported the market had reached $9.4 billion in annual revenue in China alone.

Owl&Co estimates the global vertical video economy will generate $150 billion in revenue (excluding China) in 2026 and AI will further change the economics. Before AI video, an 80-episode micro-drama typically cost 1.4 million to 2 million yuan, or about $200,000 to $280,000, and took three to four months to produce with a crew of 20 to 40 people. With tools such as Seedance 2.0, comparable series can now be made for 50,000 to 100,000 yuan, or roughly $7,000 to $14,000, in less than a month. In some cases, a solo creator with a script and AI tools can produce a competitive series.

That matters because Chinese consumers had already proven they were willing to pay for serialized mobile drama. ByteDance was not testing video AI in a lab. It was deploying it into a market already worth hundreds of billions of yuan.

Western video AI labs have had no equivalent proving ground. In the West, sophisticated media buyers tend to move slowly, negotiate carefully and represent a relatively small number of large contracts. ByteDance, by contrast, can sell into an ecosystem of hundreds of thousands of micro-drama studios serving hundreds of millions of daily viewers.

4. Supporting Video AI is An Government Industrial Policy

Local governments in Shenzhen and Shanghai have identified AI-generated micro-dramas as part of a broader industrial strategy to dominate the multibillion-dollar digital entertainment market, offering state grants of up to 2 million yuan, or about $275,000, for individual productions.

China’s National Development and Reform Commission has also included video-generation infrastructure in its AI industrial-base funding program.

Runway’s investors include General Atlantic, Google, Nvidia and Salesforce. China’s leading video AI labs, by contrast, are backed by sovereign and state-linked capital with a level of patience and time horizon that few venture funds can match.

5. The Open-Source Flanking Strategy

Meta pioneered open-weight large language models with Llama as a strategic counter to OpenAI and Google. Alibaba appears to be running a similar play in video.

Alibaba’s Wan 2.7 is open weight, with its model parameters publicly available. That is a deliberate strategy. By releasing capable models openly, Chinese labs can accelerate global developer adoption and position their architectures as the default reference point for the next generation of fine-tuned models. It also builds goodwill with the creator community and puts pressure on the revenue model of closed Western competitors that rely on charging for API access.

Sora was ultimately undone by a deeper structural problem in the U.S. Video generation at scale is a money-losing business if your only revenue model is subscriptions from individual creators. It becomes a real profit center only when you also own the platform that monetizes the content. That is what OpenAI tried to do, but failed miserably.

ByteDance, by contrast, owns Douyin, TikTok and RedNote’s Red Fruit-style distribution channels. Google could potentially pursue a similar strategy, but it remains constrained by far more intense U.S. intellectual-property scrutiny.

The result is a dual content universe. Chinese labs are winning the mass-market creator-economy and micro-drama segments through price and vertical integration. Google and Runway are positioned to win in the enterprise, agency and Hollywood segments, where compliance, provenance and creative control matter more. Two markets, two technology stacks and two very different incentive systems now coexist uneasily, as so much of U.S.-China tech competition does today.