Nvidia is building Nemotron 4, a new open-weight AI model family whose largest model will have at least one trillion parameters. The move, first reported by The Information, aims to put the chipmaker among the best freely available models in the world. Yet even at that scale, Nvidia would only be reaching a size that Chinese labs already occupy.
The earliest possible release for Nemotron 4 is fall 2026. The model family is twice the size of Nemotron 3 Ultra, which launched in June 2026. At that time, Nemotron 3 Ultra was the strongest open US model on the Artificial Analysis Intelligence Index, but it still trailed Moonshot AI's Kimi K2.6.
A Scale Gap With Chinese Labs
Nvidia's trillion-parameter ambition lags behind what Chinese labs have already shipped. Moonshot AI's Kimi K3 has 2.8 trillion parameters. DeepSeek's DeepSeek V4 Pro has 1.6 trillion parameters. Both far exceed the planned size of Nemotron 4's largest model.
The performance gap shows up in benchmarks too. On the current version of the Artificial Analysis Intelligence Index, Nemotron 3 Ultra scores 38 points. Kimi K3 scores around 60. That is a wide margin for a company that sells the very chips used to train these models.
Spending Rises as Competition Intensifies
Nvidia has tripled its cloud spending on in-house model training to $28 billion through 2031. That figure covers the period from now until 2031, a long-term bet on building competitive models alongside its core GPU business. The company is primarily known as a GPU manufacturer, but it has steadily expanded into AI model development.
The spending increase signals that Nvidia sees open-weight models as a strategic necessity. Open-weight models are freely available, allowing companies to self-host them. And the more companies self-host open models, the more GPUs Nvidia sells. That logic may be the driving force behind the entire Nemotron effort.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
A Conflict With Its Own Customers
Nemotron 4 would put Nvidia in direct competition with its own major customers like OpenAI. OpenAI is one of Nvidia's largest buyers of GPUs, and it also develops proprietary models. If Nvidia ships a competitive open-weight family, it becomes both supplier and rival to the same companies.
That tension is not hypothetical. Nvidia is among the signatories of a petition against regulating open models. The Trump administration is considering targeted bans on specific Chinese models. Nvidia's opposition to regulation may be self-serving, since open models drive demand for the hardware it sells.
What Comes Next
The fall 2026 window for Nemotron 4 is still months away. By then, Chinese labs may have pushed even further ahead in scale. The race is moving fast, and Nvidia's entry into the trillion-parameter club may already be late.
Jonathan Kemper wrote the article for The Decoder, dated Aug 12, 2026, based on the original report from The Information. The story highlights a simple reality: Nvidia wants to lead in open models, but its scale ambition trails the frontier set by Moonshot AI and DeepSeek.