Nvidia’s Nemotron 4 targets trillion-parameter scale
Nvidia is developing Nemotron 4, an open-weight model family intended to compete with leading freely available AI systems. According to The Information, its largest version is planned to contain at least one trillion parameters, twice as many as Nemotron 3 Ultra. Nvidia has also tripled its planned cloud spending for internal model training, reaching $28 billion through 2031. The earliest possible release is this fall.
That target would match a scale already reached by Chinese laboratories. Moonshot AI’s Kimi K3 has 2.8 trillion parameters, while DeepSeek V4 Pro has 1.6 trillion. Nemotron 3 Ultra, released in June, was the strongest open US model on the Artificial Analysis Intelligence Index at launch, although it ranked below Kimi K2.6. In the index’s current version, Nemotron 3 Ultra scores 38 points and Kimi K3 about 60.
Nvidia has signed a petition opposing regulation of open models as the Trump administration considers targeted bans on particular Chinese models. Wider use of self-hosted open models could increase demand for Nvidia GPUs, while Nemotron 4 would also compete directly with major customers such as OpenAI.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.