Chinese AI Labs Now Lead the US in Video Generation

Chinese artificial intelligence labs have pulled ahead of their American rivals in one of the hottest areas of generative AI: video generation. Companies like ByteDance and Kuaishou are leveraging enormous libraries of short-form video content from their own platforms to train models that now outperform anything coming out of the US. The gap is real, and developers working in the space say it is widening.

A Data Advantage the US Cannot Match

The key factor behind China's lead is training data. ByteDance, the parent company of TikTok, and Kuaishou, which runs one of China's largest short-video apps, sit on massive collections of user-generated video clips. These datasets give their AI models a structural advantage that US labs simply do not have access to. Training a video generation model, which is a type of large language model adapted for visual content, requires enormous volumes of diverse, high-quality video. Chinese labs also employ film students to label video data using professional cinematic vocabulary, producing training annotations that are far richer than typical engineering-first approaches. This combination of scale and labeling quality creates a data moat that is extremely difficult for Western competitors to replicate.

Top Benchmarks Tell the Story

The results speak for themselves. On the Artificial Analysis Video Arena, a widely used leaderboard for ranking AI video generators, the top four spots are all held by Chinese models. Alibaba's RynnBrain, ByteDance's Seedance 2.0, SkyReels V4, and Kuaishou's Kling 3.0 all rank above the highest-placed Western model. Meanwhile, OpenAI shut down its own video tool Sora in April, citing high compute costs and a strategic shift toward coding tools and artificial general intelligence research. Kuaishou's Kling alone has attracted more than 60 million users worldwide, showing strong consumer demand for these tools well beyond China's borders.

The shift carries broad implications for the global AI landscape. While the US still leads in raw computing power thanks to advanced chips from companies like Nvidia, China's advantage in video data and labeling quality could prove difficult to close. For businesses and creators exploring AI-powered video tools, the most capable options increasingly come from the east. This marks a notable reversal in a field where American companies have long set the pace for innovation.