# Nvidia 打造 Nemotron 4，目标万亿参数规模

- 来源：The Decoder：AI News（RSS）
- 作者：Jonathan Kemper
- 发布时间：2026-08-12 20:37
- AIHOT 分数：55
- AIHOT 链接：https://aihot.virxact.com/items/cmsq34rrm01twro2eknem7d48
- 原文链接：https://the-decoder.com/nvidias-nemotron-4-aims-for-one-trillion-parameters-a-scale-chinese-labs-already-surpassed

## AI 摘要

Nvidia 正在开发开源权重模型 Nemotron 4，其最大版本参数至少达一万亿，为 Nemotron 3 Ultra 的两倍，最早可能于今年秋季发布。为此，Nvidia 已将内部模型训练的云支出增至 280 亿美元，持续至 2031 年。

## 正文

Nvidia is building Nemotron 4, a new open-weight model designed to compete with the best freely available models in the world. According to The Information, the largest model in the family will have at least one trillion parameters, twice the size of Nemotron 3 Ultra. Nvidia has tripled its cloud spending on in-house model training to $28 billion through 2031. The earliest possible release would be this fall.

Even at one trillion parameters, Nvidia would only be reaching a scale that Chinese labs already occupy. Moonshot AI's Kimi K3 has 2.8 trillion parameters, and DeepSeek V4 Pro has 1.6 trillion. At launch in June, Nemotron 3 Ultra was the strongest open US model on the Artificial Analysis Intelligence Index, but still trailed Kimi K2.6. On the current version of the index, Nemotron 3 Ultra scores 38 points and Kimi K3 around 60.

Nvidia is also among the signatories of a petition against regulating open models, even as the Trump administration considers targeted bans on specific Chinese models. The more companies self-host open models, the more GPUs Nvidia sells. But Nemotron 4 would also put Nvidia in direct competition with its own major customers like OpenAI.