NVIDIA Becomes an AI Open Model Maker

NVIDIA has built much of its recent success supplying chips to companies developing artificial intelligence systems. With the release of its Nemotron 3 models, the company is taking a step toward becoming a more visible model maker.

The announcement includes a series of advanced open AI models, along with training data details and tools intended to help engineers download, modify, and run the models on their own hardware. According to NVIDIA, the release is designed to support developers building agentic AI systems at scale.

A Shift in a Changing AI Industry

The release comes as major AI companies such as OpenAI, Google, and Anthropic are increasingly developing their own chips. WIRED notes that this trend could reduce long-term reliance on NVIDIA’s hardware, even as those firms remain major customers today.

At the same time, open models continue to play a significant role in the AI ecosystem. Researchers and startups use them to experiment, prototype, and build systems without relying exclusively on closed APIs. While US companies offer some open models, they tend to release and update them less frequently than their Chinese competitors, according to data cited by WIRED from the open-source hosting platform Hugging Face.

As a result, open models from Chinese companies have become especially popular for experimentation and development.

What NVIDIA Is Releasing

Nemotron 3 consists of multiple model variants. NVIDIA announced three sizes: Nano, Super, and Ultra. The Nano version is available first, while the larger models are intended for later release.

The models are designed to be downloaded and run on private infrastructure. NVIDIA has shared benchmark scores indicating that Nemotron 3 ranks among the strongest open models currently available, according to the company.

NVIDIA is also releasing information about the data used to train the models. This level of transparency is broader than that offered by many US-based AI companies and is intended to make the models easier to modify and fine-tune.

In addition, NVIDIA has introduced tools to support customization, including a hybrid latent mixture-of-experts architecture. According to NVIDIA, this architecture is particularly suited to building AI agents capable of taking actions on computers or the web. The company is also launching libraries that support reinforcement learning, enabling developers to train agents using simulated rewards and penalties.

Open Models and Agentic Systems

NVIDIA executives have emphasized the importance of open models for AI builders. Kari Ann Briski, Vice President of Generative AI Software for Enterprise at NVIDIA, outlined three reasons open models matter. She pointed to the need to customize models for specific tasks, the value of offloading queries to different models, and the ability to improve reasoning through post-training techniques.

“Open source is the foundation for AI innovation,” Briski said, adding that such approaches continue to accelerate adoption across industries .

CEO Jensen Huang reinforced that message in NVIDIA’s announcement, saying open innovation drives AI progress. He added that Nemotron provides developers with greater transparency and efficiency when building agentic systems.

A Broader Industry Trend

The Nemotron release reflects a broader pattern in the AI industry. Over the past year, US firms have become more guarded about their research and engineering practices, sharing fewer technical details as competition has intensified.

At the same time, open models have continued to see widespread use. A report from OpenRouter, cited by WIRED, found that open models accounted for around a third of the tokens processed through its systems in 2025. Chinese companies such as DeepSeek, Alibaba, Moonshot AI, Z.ai, and MiniMax regularly release open-source models and publish research details. These practices have helped them gain popularity among developers.

Geopolitical Considerations

WIRED also notes that these trends intersect with geopolitical pressures. NVIDIA’s hardware has become significant enough to factor into US-China trade discussions. The US government recently signaled that it would allow NVIDIA to export its H200 chips to China. At the same time, China has moved to encourage domestic alternatives and reduce reliance on foreign technology.

This dynamic could lead AI models to become more closely aligned with specific hardware ecosystems, according to the article, potentially posing challenges for NVIDIA’s position over time.

What the Release Represents

WIRED characterizes NVIDIA’s move as notable because it places the company more directly in the model-making landscape, even as it continues to dominate the hardware layer. The article suggests that NVIDIA’s support for open-source AI may be tied to broader shifts in how models and chips are developed together.

As WIRED summarizes, the world’s leading chipmaker “wants open source AI to succeed – perhaps because closed models increasingly run on its rivals’ silicon”.