Nvidia discusses $2,000 annual GPU fee for AI agents

By Grace Kim

Nvidia CEO Jensen Huang put annual licensing for software that runs AI agents at about $2,000 per graphics processing unit (GPU) on October 22 at GTC Berlin, according to The Register. The figure could create recurring software revenue at infrastructure scale, but Nvidia has not confirmed a start date, standard price or package.

At GTC Berlin, Nvidia outlined its push across chips, software and AI-agent security. Huang said: “Software licensing will become increasingly a major part of our company’s business. The way that we license will be…for example, it’s about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories.”

His “AI factories” are Nvidia’s term for the computing infrastructure hosting those agents.

What does the $2,000 fee change commercially?

A FourWeekMBA scenario assumes paid licenses for five million GPUs. At $2,000 per GPU per year, Nvidia would generate $10 billion annually. If all five million customers renewed for five years, the gross total would reach $50 billion.

For Nvidia, the model would create recurring software revenue from licensed GPU capacity and widen its infrastructure business beyond hardware. For enterprise buyers, however, no firm pricing or package follows yet: Nvidia has disclosed neither the number of GPUs covered nor a start date, and has not defined which software the fee includes.

Nvidia already lists its AI Enterprise software platform at $4,500 per GPU for a one-year subscription, more than twice the figure discussed by Huang. The two prices are not directly comparable because Nvidia has not said whether they cover the same products or licensing terms. The $2,000 proposal therefore is not a confirmed price cut.

The infrastructure behind agent software

AI agents can use tools, access data and execute tasks, creating security and management problems when enterprises run them at scale. On October 21, Nvidia announced the OpenShell Reference Architecture, a system design that combines:

• Vera Rubin computing infrastructure.
• Vera CPUs.
• BlueField-4 data processing units (DPUs) running Nvidia Sentry.

OpenShell provides a secure runtime (a controlled execution environment) for AI agents. It establishes boundaries around their actions and access to tools and resources. According to Nvidia, Sentry monitors agent activity independently and can quarantine an agent that tries to cross those boundaries.

The design puts security controls in the underlying infrastructure rather than relying only on safeguards built into an AI model or application.

Nvidia introduced the Open Agent Safety Platform in September. The newer reference architecture builds on that effort and extends the approach to enterprise infrastructure and edge systems.

Other companies are incorporating those controls into their products. Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents. The setup gives businesses more control over what AI agents can access inside execution sandboxes (isolated agent environments).

Salesforce has developed a Slack Admin Bridge for OpenShell. It lets administrators review agent-access requests, approve or reject permissions and monitor activity directly in Slack.

The wider business strategy

The proposed pricing fits Nvidia’s wider argument that it is becoming more than a GPU supplier. The company has expanded into networking, systems, enterprise software and infrastructure designed for AI workloads.

On the BG2Pod podcast, Huang said: “One of these days, people might not categorize us as a chip company anymore.” He cited Nvidia’s presence in healthcare, automotive, telecommunications and life sciences, where the company offers broader technology stacks rather than chips alone.

Nvidia still has not said when the fee would take effect, whether $2,000 would be standard, which software it would cover or how it would fit with AI Enterprise. Until those terms are published, the number marks a possible pricing model rather than a confirmed commercial schedule.