$2,000 per GPU per year
Speaking at GTC Berlin on October 22, NVIDIA CEO Jensen Huang said the company envisions an annual license fee of about $2,000 per GPU for the software that runs agents on its AI factories. Huang said software licensing will become an increasingly major part of the company's business.
The math and the unknowns
At that rate, licensing five million GPUs would generate $10 billion a year, or $50 billion over five years. However, NVIDIA has not disclosed how many GPUs the $2,000 fee would cover, whether it is a standard price, or which software products it includes. The company's existing AI Enterprise platform is listed at $4,500 per GPU for a one-year subscription; it is unclear whether the two prices cover the same products, so the figure should not be treated as a confirmed price cut.
Agent security infrastructure
The licensing discussion comes as NVIDIA builds out the infrastructure needed to deploy and manage AI agents in enterprise environments. On October 21, the company announced its OpenShell Reference Architecture, which combines Vera Rubin computing infrastructure, Vera CPUs and BlueField-4 DPUs running NVIDIA Sentry. OpenShell provides a secure runtime that sets boundaries around agents' actions and access to tools, while Sentry independently monitors agent activity and can quarantine an agent that attempts to cross those boundaries.
Ecosystem grows
NVIDIA introduced its Open Agent Safety Platform in September; the newer reference architecture extends that effort to enterprise infrastructure and edge systems. Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents to give businesses more control over what their agents can access inside execution sandboxes. Salesforce has developed a Slack Admin Bridge for OpenShell that lets administrators review agent access requests, approve or reject permissions, and monitor activity directly in Slack.
