China Now Makes Almost Every Top AI Video Model

The most striking sign of China's progress in artificial intelligence right now is not a chatbot or a coding tool. It is video. On the public leaderboard kept by the benchmarking firm Artificial Analysis, nine of the ten highest-rated text-to-video systems are made by Chinese companies. The only non-Chinese entry near the top belongs to Google. Every other name in the ranking comes from Beijing, Hangzhou or Shenzhen.

That is the picture Bloomberg laid out in an August 9 analysis titled "Chinese AI Video Is Coming for More Than Hollywood." The piece argued that the clearest evidence of Chinese capability is no longer in language models, where American labs still hold an edge, but in generative video, where Chinese firms have quietly pulled ahead.

Who is on the board

The names will be unfamiliar to most Western readers, which is part of the story. Kuaishou's Kling, MiniMax's Hailuo, ByteDance's Seedance and Alibaba's Wan trade places at the top of the rankings, and each has shipped rapid upgrades over the summer. In late July, MiniMax and ByteDance released competing video models within days of each other, the kind of head-to-head cadence that has become normal in China and rare elsewhere.

These systems turn a short written prompt into a few seconds of coherent, physically plausible footage. A year ago the results were dreamlike and unusable. Now they are close enough to real that advertising agencies, game studios and short-video platforms have started to fold them into ordinary production.

Why video, and why now

Part of the answer is structural. China has enormous short-video platforms in Douyin and Kuaishou, which means both the training data and the paying customers sit inside the same companies building the models. Part of it is strategy. With American export controls limiting access to the most advanced chips, Chinese labs have leaned into areas where cleverness with data and architecture matters as much as raw compute, and video has rewarded that approach.

The gap is not total. American analysts are quick to point out that the United States still leads on the frontier language and reasoning models that anchor most enterprise work, and that Google's Veo line remains competitive at the very top of the video charts. A CNBC report on August 7 made the case that the country retains a real advantage in the models businesses actually deploy. But the direction of travel in video is hard to argue with.

This fits a wider pattern we have been tracking. Open-weight Chinese models like Alibaba's Qwen 3.8 Max and DeepSeek's V4 Flash have been closing on their American rivals for months, and Washington has begun to weigh curbs on Chinese open models as they gain ground abroad.

What it means beyond Hollywood

The Bloomberg headline is pointed for a reason. Cheap, fast, controllable video generation reshapes advertising, training material, propaganda and misinformation all at once. When the best tools for making synthetic footage sit largely in one country's hands, that becomes a question of soft power and security, not just a creative one. For now the leaderboard tells a simple story. In the corner of AI that turns words into moving pictures, China is not catching up. It is ahead.