NVIDIA is accelerating the construction of its AI ecosystem moat with a "hardcore open source" strategy. This Monday, the GPU giant announced two key initiatives: on one hand, acquiring SchedMD, the developer of the globally mainstream high-performance computing job scheduling system Slurm; on the other hand, launching a brand-new open-source large model family called Nemotron 3, fully betting on the next wave of AI agents (Agentic AI) and physical AI (Physical AI).

At the infrastructure level, NVIDIA has officially incorporated Slurm into its portfolio. Since its inception in 2002, Slurm has become the de facto standard scheduling tool for global supercomputing centers and AI clusters, managing vast computing resources including those in the world's Top500 supercomputers. SchedMD was founded in 2010 by Morris Jette, the core developer of Slurm, and current CEO Danny Auble. The company has had more than a decade of collaboration with NVIDIA. After the transaction, NVIDIA has promised that Slurm will continue to operate in an open-source, vendor-neutral manner and will increase investment to "accelerate its integration into various systems." This move not only strengthens NVIDIA's control over the AI infrastructure software stack but also ensures optimal support for its GPUs at the scheduling layer—laying the groundwork for future large-scale AI clusters.

At the model level, NVIDIA launched the open-source Nemotron 3 model family, claiming it to be the most efficient open-source model series for building high-precision AI agents. This family includes three models tailored for different scenarios:

- Nemotron 3 Nano: A lightweight model suitable for edge devices or specific tasks;

- Nemotron 3 Super: Designed for multi-agent collaborative systems, supporting complex task decomposition and collaboration;

- Nemotron 3 Ultra: For high-complexity reasoning tasks, with stronger logical and planning capabilities.

NVIDIA CEO Jensen Huang emphasized, "Open innovation is the cornerstone of AI progress. Through Nemotron, we will transform advanced AI into an open platform, providing developers with the transparency and efficiency needed to build scalable agent systems."

This series of actions are not isolated. Last week, NVIDIA also released an open-source vision-language model for autonomous driving research called Alpamayo-R1, and expanded the developer documentation and workflow support for its open-source "world model" Cosmos. These measures all point to a strategic focus: Physical AI—that is, AI systems capable of perceiving, reasoning, and acting in the physical world, such as robots and autonomous vehicles.

NVIDIA is trying to become the "full-stack supplier" of the Physical AI era: from GPU hardware, Slurm scheduling systems, Cosmos world models, to Nemotron agent models, forming a closed-loop ecosystem. While competitors are still vying for the discourse on general large models, NVIDIA has quietly extended the battlefield to a new dimension: how AI interacts with the real world.