Nvidia's Huang pegs AI software at about $2,000 per GPU a year
Speaking at GTC Berlin, the Nvidia CEO said the fee would cover software for running agents on AI factories; the company has not set a launch date.
Nvidia CEO Jensen Huang said licensing Nvidia's AI software to run agents on AI factories could cost about $2,000 per GPU per year, a figure he discussed at GTC Berlin on October 22, according to tomshardware.com.
Huang said the roughly $2,000-per-GPU annual fee would cover software Nvidia provides for running agents on AI factories. Nvidia's AI Enterprise platform, by comparison, is listed at $4,500 per GPU for a one-year subscription.
Nvidia has not disclosed how many GPUs the proposed $2,000 fee would cover, or whether it represents a standard price. It is also unclear which software the proposed license would cover, or whether it overlaps with other Nvidia products. The company has not said when the fee would take effect.
The pricing remarks follow Nvidia's push into agent software. On October 21, the company announced its OpenShell Reference Architecture, combining Vera Rubin infrastructure, Vera CPUs and BlueField-4 DPUs running Nvidia Sentry. In September, Nvidia introduced its Open Agent Safety Platform.
Partners have also built on the stack: Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents, and Salesforce developed a Slack Admin Bridge for OpenShell. Separately, on the BG2Pod podcast, Huang said people might not categorize Nvidia as a chip company anymore.
Quick answers
How much would Nvidia's AI software cost per GPU?
Jensen Huang put the annual cost at about $2,000 per GPU per year, discussed at GTC Berlin on October 22.
When would the $2,000-per-GPU fee take effect?
Nvidia has not said when the fee would take effect.
What does Nvidia's AI Enterprise subscription cost?
Nvidia's AI Enterprise platform is listed at $4,500 per GPU for a one-year subscription.