Nvidia is preparing a new open-weight AI model family called Nemotron 4, and the scale target is significant: the largest model is expected to have at least one trillion parameters. The move would place Nvidia more directly in the race for top freely available models, while also sharpening its position in a market where Chinese labs have already pushed beyond that size.
What Nvidia is building
According to The Information, Nemotron 4 is being designed to compete with the strongest freely available models in the world. The largest model in the family will have at least one trillion parameters, which would make it twice the size of Nemotron 3 Ultra.
That matters because parameter count is one of the easiest signals to understand in the AI model race, even if it does not tell the whole story. A larger model can carry more capacity, but rankings and real-world usefulness also depend on training quality, architecture, data, deployment constraints, and how the model behaves in practical tasks.
The reported investment behind the effort is also notable. Nvidia has tripled its cloud spending on in-house model training to $28 billion through 2031. The earliest possible release for Nemotron 4 would be this fall.
Why one trillion parameters may not be enough to lead
A one trillion-parameter Nemotron 4 would be a major step up from Nemotron 3 Ultra, but it would not set the global scale ceiling. Chinese labs already operate at or above that territory.
Moonshot AI's Kimi K3 has 2.8 trillion parameters. DeepSeek V4 Pro has 1.6 trillion. By that measure, Nvidia would be entering a scale range that competitors have already reached rather than opening a new one.
The gap is not only about model size. At launch in June, Nemotron 3 Ultra was the strongest open US model on the Artificial Analysis Intelligence Index, but it still trailed Kimi K2.6. On the current version of the index, Nemotron 3 Ultra scores 38 points, while Kimi K3 scores around 60.
Those figures frame the challenge for Nemotron 4. Doubling the size of Nemotron 3 Ultra could help Nvidia close distance, but the source data does not show that parameter growth alone would be enough to overtake the models now leading the index.
The open-weight strategy has business logic
Nvidia's interest in open-weight models is closely connected to its broader hardware business. The more companies self-host open models, the more GPUs Nvidia sells. Open models can encourage organizations to run AI systems on their own infrastructure, which creates demand for the compute needed to train and serve them.
That gives Nvidia a direct incentive to support a strong open-model ecosystem. If Nemotron 4 becomes a serious option for companies that want freely available models, it could help expand the market for GPU-backed deployment.
At the same time, this strategy creates tension. Nemotron 4 would also put Nvidia in direct competition with major customers like OpenAI. Nvidia supplies the hardware that many AI developers rely on, but a stronger in-house model family makes Nvidia more than an infrastructure provider.
Politics complicates the open model race
The competitive picture is not only technical. Nvidia is among the signatories of a petition against regulating open models, while the Trump administration considers targeted bans on specific Chinese models.
That creates a difficult backdrop for open AI development. Companies want access to powerful freely available models, but governments are weighing limits on particular systems. In that environment, a US open model from Nvidia could carry strategic importance beyond benchmarks.
Still, the source facts point to a clear near-term reality: Nvidia is trying to scale up quickly, but Chinese labs have already built larger open models and are scoring higher on the cited index. Nemotron 4 may narrow the gap, yet it enters a field where scale, rankings, policy, and hardware economics are all moving at once.
What to watch next
The key question is not simply whether Nemotron 4 reaches at least one trillion parameters. The bigger test is whether it can improve meaningfully on Nemotron 3 Ultra's position against models such as Kimi K3, DeepSeek V4 Pro, and Kimi K2.6.
Several signals will matter when the model appears:
- Whether the largest Nemotron 4 model launches at the reported one trillion-parameter scale or above it.
- How it performs on the Artificial Analysis Intelligence Index compared with Nemotron 3 Ultra's 38 points and Kimi K3's score of around 60.
- Whether companies treat it as a practical open-weight option for self-hosting.
- How Nvidia balances open model ambitions with its relationships with major customers like OpenAI.
For now, Nemotron 4 looks less like a quiet research project and more like Nvidia's attempt to compete in the open-weight model race from both sides: as a supplier of the GPUs that power the field and as a builder of models meant to run on them.