Nvidia Builds Trillion-Parameter Nemotron 4 AI Family
The chipmaker is developing a new open-model family as Chinese systems intensify pressure on U.S. AI labs.
Topics
News
- Temasek, Seraphim Back Pixxel in $100 Million Round
- UN Rights Chief Warns Advanced AI Could Threaten Humanity
- Tata-owned JLR to Cut About 4,000 Bobs Amid Tariff, China Pressure
- Anthropic Walks Away From $6 Billion Decart Deal
- Hyundai India Targets 20% Women in Executive Workforce by 2030
- Security Breach Drains $320 Million in Bitcoin From Liquid Network
Nvidia is developing a Nemotron 4 model with at least 1 trillion parameters, pushing the chipmaker deeper into the race to build powerful open AI systems.
The largest model is still being trained and could be ready as early as late fall, The Information reported, citing employees working on the project. Nvidia has not announced a release date or confirmed the parameter count.
The effort expands Nvidia’s role beyond selling the chips that underpin much of the AI industry. While OpenAI, Anthropic and Google largely keep their most capable models closed, Nvidia has continued investing in models that developers can access and customize.
Nvidia had already announced the Nemotron 4 family in March when it formed the Nemotron Coalition with companies including Mistral AI, Perplexity, Cursor and Sarvam. The first model from the group, being co-developed by Nvidia and Mistral, is intended to underpin the Nemotron 4 family.
The reported scale would mark Nvidia’s most ambitious model effort yet and extend its role beyond supplying the chips and computing infrastructure used to train many leading AI systems.
“Nvidia is investing in Nemotron because we believe every company and every country needs accessible frontier open models to strengthen safety and security, accelerate innovation, and provide a foundation they can rely on from one generation to the next,” Kari Briski, Nvidia’s vice president of generative AI, said in a statement to Reuters.
Open-weight models have gained attention as companies look for alternatives to expensive proprietary AI services and Chinese developers narrow the performance gap with leading US systems.
Models from companies including Moonshot AI, Alibaba and DeepSeek have also increased competitive pressure on American developers.
Open-weight systems make their trained model parameters available for users to download, run and modify, but do not necessarily disclose training data, code or the full development process required to qualify as fully open source.
Nvidia has also become more vocal about keeping the US competitive in this market. Last month, it joined other technology companies in supporting wider development of open-weight AI and formed an industry coalition focused on AI safety and cybersecurity tools.
Separately, Nvidia on Tuesday introduced Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model aimed at tasks including code review, tool use, security monitoring and customer-service queries.
It also released NeMo Switchyard, an open-source routing system that directs individual AI tasks to the model best suited to handle them based on factors such as capability, latency and cost.
If the reported trillion parameter version reaches release, Nemotron 4 would put the company that supplies much of the computing behind frontier AI into more direct competition with the labs building the models themselves.
