NVIDIA's Nemotron 4 aims for 1 trillion parameters, still trails China

1 trillion parameters, but not even second place

The largest version of NVIDIA's next-generation open-weight model, Nemotron 4, is being designed with at least 1 trillion parameters, according to The Information. That's double the size of Nemotron 3 Ultra, released last June. But the number doesn't necessarily put it at the top of the industry. Moonshot AI's Kimi K3 reportedly has 2.8 trillion parameters, and DeepSeek's V4 Pro has 1.6 trillion. Even if NVIDIA hits the 1 trillion mark, it would only be catching up to a scale Chinese labs have already surpassed.

A performance gap remains too

Parameter count alone doesn't guarantee performance, but the current benchmark gap is clear. Nemotron 3 Ultra was the strongest open-source American model at launch, but it fell short of Kimi K2.6, which held the top spot at the time. In the latest version of the Artificial Analysis Intelligence Index, Nemotron 3 Ultra reportedly scored 38, while Kimi K3 scored around 60. That gap of more than 20 points is what Nemotron 4 will need to close.

Model	Parameters	Intelligence Index	Developer
Nemotron 4 (upcoming)	1 trillion+	Not disclosed	NVIDIA
Kimi K3	2.8 trillion	60	Moonshot AI
DeepSeek V4 Pro	1.6 trillion	Not disclosed	DeepSeek
Nemotron 3 Ultra (current)	Half of Nemotron 4	38	NVIDIA

Why NVIDIA is pouring in $28 billion

NVIDIA has reportedly tripled its cloud investment for training its own models to $28 billion through 2031. A company that sells GPUs is now also building large language models itself, and there's a business logic behind it: the more companies run open models on their own servers, the more GPU demand grows. As our publication reported on August 8, NVIDIA has backed the spread of open-weight models, including by signing the "Open Weights and American AI Leadership" open letter last July alongside more than 200 companies and institutions. The physical AI model Cosmos 3, also unveiled around the same time, is part of this same open-model strategy.

Caught between regulation and its own customers

At the same time, NVIDIA has also signed a petition opposing regulation of open models. This comes as the Trump administration reportedly considers restrictions targeting specific Chinese models. The issue is that the more successful Nemotron 4 becomes, the more directly NVIDIA ends up competing with major customers like OpenAI. Selling hardware while also selling models built on that hardware could send a delicate signal to its own customers.

So what changes

Nemotron 4 is reportedly expected to launch as early as this fall. Once it's out, benchmark performance is likely to matter more than parameter count as the key measure. This news shows that a trilion-parameter scale alone no longer guarantees a top ranking. It suggests the open-weight race between U.S. and Chinese labs is shifting from a battle over scale to one over closing the actual intelligence-score gap. For NVIDIA, it remains to be seen how long its two business lines — selling GPUs and building its own models — can coexist without friction.