Nvidia is quietly working on a massive new release to expand its footprint in artificial intelligence. According to internal sources, the graphics processing giant is actively training Nemotron 4, a new family of open-weight models designed to take on the world’s most powerful open systems.
The flagship variant in the lineup is expected to reach at least 1 trillion parameters, roughly doubling the scale of the company’s previous model releases. While final training is still underway and Nvidia has yet to set an official release date, employees involved in the project suggest it could land as early as late fall.
The push into high-parameter open software highlights a strategic shift for the chip maker. Nvidia already dominates the hardware supply for AI training and inference, but building top-tier models gives the firm direct influence over software development. By releasing open weights, Nvidia provides enterprise clients and sovereign nations with alternatives to closed commercial application programming interfaces (APIs), giving companies a reason to host heavy workloads directly on Nvidia hardware.
The move comes at a competitive moment in tech. Open-source models from international labs have surged in capability, narrowing the gap with closed systems from leading developer labs. By doubling down on the Nemotron lineup, Nvidia aims to deliver a reliable foundation for enterprise tasks like complex reasoning, agentic tool use, and automated coding.
Addressing the strategic direction, Kari Briski, vice president of generative AI at Nvidia, emphasized that accessible, open-frontier models are essential for individual countries and companies to strengthen security and drive local innovation over time. If delivered as planned, Nemotron 4 could set a strong new baseline for accessible enterprise hardware setups later this year.
kanu priya
