The History of AI

China and the chips

The United States restricts sales of advanced chips to China, and Chinese labs respond with efficient open models and, later, with chips of their own.

  1. Chip export controls on China (October 7, 2022)

    The United States restricts sales to China of advanced chips and the equipment to make them. Compute becomes a matter of national security.

  2. DeepSeek-V3 (December 26, 2024)

    DeepSeek trains V3 on H800 GPUs, the version Nvidia was allowed to sell in China, and with far less compute than usual.

  3. DeepSeek-R1 (January 20, 2025)

    R1 arrives on top of V3, an open reasoning model very close to the best American ones. Nvidia's stock falls 17 percent.

  4. Qwen3 (April 29, 2025)

    Alibaba releases Qwen3, a family of open hybrid reasoning models in every size.

  5. Kimi K2 (July 11, 2025)

    Moonshot AI releases Kimi K2, an open model with a trillion parameters built to work as an agent. The open frontier is now mostly Chinese.

  6. DeepSeek V4 (April 24, 2026)

    DeepSeek prepares V4 to run on Huawei's Ascend processors, China's alternative to the chips the controls restrict.

Open in the interactive timeline