Nvidia plans to release Nemotron 4, an open-weight model with at least one trillion parameters, by this autumn. This new system would double the size of the previous Nemotron 3 Ultra and requires a significant increase in internal cloud spending to $28 billion through 2031. The target scale is already occupied by Chinese laboratories, where Moonshot AI’s Kimi K3 contains 2.8 trillion parameters and DeepSeek V4 Pro holds 1.6 trillion. Nemotron 3 Ultra currently scores 38 points on the Artificial Analysis Intelligence Index, while Kimi K3 scores around 60. Nvidia has also joined a petition opposing regulations on open models despite potential bans on specific Chinese systems from the Trump administration.
The move places Nvidia in direct competition with its own major customers, including OpenAI, while the company benefits from increased GPU sales as organisations self-host these models. The scale race highlights a gap between American and Chinese development priorities, with US firms focusing on open weights while Chinese labs prioritise massive parameter counts. Nvidia’s strategy relies on the assumption that larger models will drive hardware demand even as performance gaps narrow.
* Release expected this fall
* Internal cloud spending tripling to $28 billion
* Open-weight design targets freely available models
