NVIDIA Expands Nemotron, Introduces Switchyard Router
Nvidia unveiled Nemotron 3.5 Lightning, a lightweight open-source 30B MoE model that can run on a single GPU and is designed for fast, customizable local AI agents and task-specific use cases, with open weights enabling fine-tuning. The company also open-sourced cuFile APIs and the supporting storage stack to streamline direct GPU access in AI pipelines, reflecting its broader open AI strategy and push toward AI factories as CEO Jensen Huang defends open models.