The Information reported that Nvidia is developing Nemotron 4, a family of open-source language models, whose largest version should have at least 1 trillion parameters — twice the size of Nemotron 3 Ultra.
The earliest possible launch would be in the Northern Hemisphere fall. Nvidia tripled its cloud computing spending for internal model training, reaching US$ 28 billion by 2031.
Scale still below Chinese models
Even at the 1 trillion parameter scale, Nvidia would reach a level that Chinese labs have already surpassed. Moonshot AI's Kimi K3 has 2.8 trillion parameters, and DeepSeek V4 Pro has 1.6 trillion.
At the launch of Nemotron 3 Ultra, in June, the model was the strongest among US open models on the Artificial Analysis Intelligence Index, but fell behind Kimi K2.6. In the current version of the index, Nemotron 3 Ultra scores 38 points, while Kimi K3 sits around 60.
Nvidia also joins the list of signatories of a petition against regulation of open models, while the Trump administration considers targeted bans on specific Chinese models. Nemotron 4, however, would put the company in direct competition with major customers, such as OpenAI.
