“We love the world where we can use both”: How Nvidia thinks about local and frontier models

Nvidia says enterprise AI will increasingly use hybrid systems that route each task to the best model.

Simple requests can go to fast local or open models, while harder problems are sent to frontier models.

Joey Conway said this setup can cut costs and latency while keeping sensitive data on-premises.

Nvidia points to enterprise collaborations showing tuned open-model systems can rival closed models on business tasks.