NVIDIA AI Releases Nemotron 3.5 Lightning: A 30B Open MoE with 3B Active Parameters, and NeMo Switchyard Model Router - MarkTechPost

NVIDIA has released two open-source AI technologies—Nemotron 3.5 Lightning and NeMo Switchyard—designed for efficient execution in agentic AI workflows. Nemotron 3.5 Lightning is a 30B parameter Mixtu... Read original →

NVIDIA’s release of Nemotron 3.5 Lightning and NeMo Switchyard marks a pivotal shift in AI inference efficiency. By integrating Mamba-2 with Mixture-of-Experts (MoE), the system achieves 30B parameters with only 3B active, leveraging Speculative Decoding and NVFP4 quantization for 4x speedup. This innovation pressures downstream model providers to optimize hybrid architectures, pushing Attention and MoE convergence as industry standards. NeMo Switchyard’s dynamic routing capability redefines AI agent deployment, accelerating multi-model orchestration. While the permissive OpenMDW-1.1 license supports broad adoption, future regulatory shifts could constrain commercialization. Competitors like Anthropic, Google, and Meta may respond with proprietary routing and inference technologies to counter NVIDIA’s infrastructure lead. In the long term, this tech will drive AI agents from lab to enterprise use cases, establishing a new AI infrastructure ecosystem centered on model orchestration.