Nvidia Unveils Dynamo 1.0 to Orchestrate AI Factories at Scale

Nvidia unveiled Dynamo 1.0, an open-source platform designed to orchestrate large-scale AI training and inference across data centers. Announced at the GPU Technology Conference in San Jose, Dynamo targets growing enterprise complexity in generative and agentic AI deployments.

The platform runs on Nvidia’s Vera Rubin NVL72 rack-scale AI supercomputer, introduced in January. It delivers up to 10× higher throughput per watt and reduces token cost by 90% compared to prior stacks.

Dynamo handles routing, caching, and scheduling for low-latency, large-context inference-critical as agentic AI evolves beyond human interaction into AI-to-AI coordination. Nvidia calls this the 'fourth scaling law.'

The platform integrates with major inference frameworks and anchors Nvidia’s new Agent Toolkit, which includes open models and microservices for building autonomous agents. Analysts say Dynamo extends Nvidia’s moat upward-shaping AI infrastructure standards through open software.