Nvidia is considering a new annual software licensing fee of approximately $2,000 per GPU for running AI agents on its infrastructure. CEO Jensen Huang discussed this potential pricing model during the GTC Berlin event held on October 22.
The company has not yet confirmed if this fee will become a standard price or which specific software products it would cover. This proposed figure is lower than the existing Nvidia AI Enterprise platform subscription, which is priced at $4,500 per GPU annually.
The initiative aligns with Nvidia's development of infrastructure for managing AI agents in enterprise environments. On October 21, the company introduced the OpenShell Reference Architecture, which integrates Vera Rubin computing, Vera CPUs, and BlueField-4 DPUs.
OpenShell utilizes Nvidia Sentry to monitor and secure AI agent activity, allowing for the quarantine of agents that exceed defined boundaries. Companies such as Anthropic and Salesforce are already integrating these controls into their own platforms.
