While the argument over China's AI progress has centered on chatbots, a quieter takeover has happened in video.
According to a Bloomberg report by Catherine Thorbecke, surfaced by Techmeme, Chinese AI labs now account for nine of the top 10 text-to-video models ranked by Artificial Analysis, the independent benchmarking outfit that scores AI systems across categories. Text-to-video models are the ones that turn a written prompt — a sentence describing a scene — into moving footage.
Bloomberg reports these models are picking up adoption globally, not just domestically, and that the lead may hand Chinese labs an edge in building so-called world models: AI systems that learn how physical environments behave, which researchers see as a foundation for robotics and other machines that have to operate in the real world.
The finding cuts against where the conversation has been. As Bloomberg notes, the debate over China's AI capabilities has been dominated by new large language models such as Moonshot's Kimi K3 — the text-based systems that grab headlines — rather than by video generation.
It also lands alongside a broader framing of the race. A Memeburn piece argues that China is closing the AI gap while US investment still sets it apart, a reminder that benchmark leadership in one category and overall financial firepower are separate scoreboards.
Why it matters: if the tools that generate video — and potentially the models that learn how the physical world works — are increasingly built in China, the assumption that American labs set the pace in generative AI needs revisiting.