SAN FRANCISCO — Nvidia is developing Nemotron 4, an artificial intelligence model expected to exceed 1 trillion parameters — roughly twice the size of Nemotron 3 Ultra, its current largest model, which launched in June with 550 billion parameters.
Development of the Nemotron line carries a budget capped at $7 billion through fiscal 2028, reflecting Nvidia's push into AI software beyond its core chip business.
The move puts Nvidia in more direct competition with its own customers. OpenAI and Microsoft both rely on Nvidia's chips to run their frontier AI models — and both now face a supplier building rival products.
Nvidia has not set a release date for Nemotron 4, but two people familiar with the project said it could arrive as early as this fall.
On Tuesday, Nvidia released Nemotron 3.5 Lightning, a smaller model with approximately 30 billion parameters. The company said it delivers up to four times faster output and completes agentic tasks 30 percent faster than comparable models. Companies can customize it with their own data, tools and workflows. Nvidia also released free model-routing software that directs tasks to different models based on cost and capability.
The open-model strategy is designed to expand the pool of optimized software running on Nvidia hardware, sustaining GPU demand even as some of the company's largest chip customers develop their own silicon — a dynamic that complicates the competitive picture on both sides.
Nvidia also announced a $500 billion financing deal with six financiers this week to fund AI infrastructure projects. The company holds a substantial investment in OpenAI, having provided $30 billion to the startup — a financial relationship that sits awkwardly alongside the emerging model competition.
Chinese AI developers continue to train on Nvidia chips despite Beijing's push to promote domestic chip alternatives, according to the South China Morning Post.
Nvidia stock closed at $217.08, down 0.2 percent.
