Nvidia Plans $26B Over 5 Years for Open-Source AI Models to Counter Chinese Models' Threat

Nvidia recently announced a major plan: it will invest $26 billion over the next five years to build open-source AI models. This move is driven by the rise of Chinese open models—DeepSeek, Alibaba Qwen, Moonshot AI, etc., have become the world's top open-weight models, widely used by overseas startups and researchers to develop applications. More crucially, industry rumors suggest DeepSeek will soon release a new model trained on Huawei chips; if true, it will prove that top models can be built without Nvidia chips, directly challenging its chip monopoly.

To counter this threat, Nvidia launched its strongest open-source model to date, Nemotron 3 Super (128 billion parameters), which scored 37 in the AI Index comprehensive rating—surpassing OpenAI's GPT-OSS (33) but still lagging behind several Chinese models. Additionally, the company has completed pre-training of a 550 billion parameter model. Nvidia executives stated that this initiative aims to help the ecosystem develop and keep developers within the Nvidia ecosystem. Nathan Lambert from the Allen Institute for Artificial Intelligence called on the U.S. government to fund open-source models.