Nvidia has released a design for running AI agents inside fixed limits, from enterprise infrastructure to edge machines. Anthropic, OpenAI, Meta, Microsoft, Google Cloud and AWS are among the companies working with it, Nvidia said.