NVIDIA has announced Spectrum-6, networking technology built for Vera Rubin and intended for what the company calls gigascale AI factories. According to the announcement, these facilities bring together hundreds of thousands of GPUs and CPUs to train frontier models, support agentic AI, and generate intelligence at large scale. NVIDIA frames networking as a critical factor in driving token generation at this level and describes Spectrum-6 as a networking milestone.
Why it matters
As the input notes, the scale of AI infrastructure has grown to systems combining hundreds of thousands of processors. In this context, the company positions networking as a power multiplier rather than a secondary concern, suggesting interconnect performance is increasingly central to large model training and deployment.
Who should care
Organizations building or operating large-scale AI data centers and those training frontier models are the primary audience indicated by the announcement.