Success isn't always about greatness. It's about consistency. Consistent hard work leads to success.

Jensen Huang Unveiled Nemotron 3 Ultra

Jensen Huang took the stage at Computex 2026 in Taipei on Sunday and unveiled Nemotron 3 Ultra, a 500 billion parameter open AI model that sits at the top of Nvidia's growing model family. The announcement positions Nvidia not just as the company that makes the hardware powering AI, but as a full AI platform company building the software layer on top of it.

The Ultra is built specifically for complex reasoning, planning, and agentic workflows, meaning AI systems that can break down multi-step tasks, execute them, and adjust along the way without needing a human to supervise every decision. It sits above two existing models in the Nemotron 3 family. The Nano handles lighter workloads, and the Super, launched in March with 120 billion parameters, targets mid-range enterprise use cases.

What makes the model technically interesting is how it handles its own scale. Rather than activating all 500 billion parameters for every task, Nemotron 3 Ultra uses a technique called latent mixture-of-experts, which routes each task only through the portions of the network that are relevant. The result is up to five times higher throughput than previous versions, which translates directly into lower cost per use at enterprise scale.

The Nemotron 3 family has already been downloaded over 50 million times in the year leading up to this launch, suggesting meaningful adoption well before the Ultra arrived. Nvidia is banking on the fact that giving developers access to powerful open models builds the kind of ecosystem loyalty that keeps them on Nvidia infrastructure for the long run.