China has achieved near-complete dominance in text-to-video AI generation, with nine of the top 10 models on independent leaderboards now originating from Chinese labs. The finding, highlighted in a Bloomberg Opinion column by Catherine Thorbecke published this week, marks a significant shift in the global AI competitive landscape.
The only non-Chinese model in the top 10 comes from Google, leaving American frontier labs—including OpenAI, Anthropic, and Meta—scrambling to close the gap in a modality that many researchers consider critical for world-model development and AI-powered content creation.
Recent releases driving the Chinese surge include ByteDance’s Seedance 2.5 and MiniMax’s H3, both of which have posted strong scores on benchmarks like Artificial Analysis’ leaderboard. The rapid improvement from multiple Chinese vendors suggests a coordinated effort across the country’s AI ecosystem, with each lab building on advances from peers.
The implications extend well beyond Hollywood-style content creation. Text-to-video capability is increasingly viewed as essential for training world models that can simulate physical environments, predict outcomes, and enable agents to reason about real-world scenarios. China’s lead in this area could provide advantages in robotics, autonomous systems, and simulation training.
US labs have focused heavily on language model reasoning and agentic capabilities, betting that these foundations would transfer to other modalities. The text-to-video gap suggests that this strategy carries risks—while American models excel at textual reasoning, Chinese advances demonstrate that multimodal leadership requires dedicated investment.
The competitive dynamic has also drawn attention from policymakers, with the video generation lead adding to concerns about China’s AI trajectory across nearly every modality. As the arms race continues, text-to-video may emerge as the most visible arena where the US technology gap with China has narrowed to single digits on independent benchmarks.