Nvidia weighs $2,000 annual GPU software fee for AI agents

By Sophie Lindqvist

Nvidia CEO Jensen Huang discussed about $2,000 per GPU per year for software running AI agents on October 22 at GTC Berlin, according to The Register. The discussion has no announced start date, defined software bundle, or disclosed number of GPUs under paid licenses.

What does the proposed fee cover?

At GTC Berlin, Huang said:

“Software licensing will become increasingly a major part of our company’s business.”

He then described an example charging about $2,000 per GPU per year for Nvidia software that runs agents on “AI factories,” rather than announcing a finalized market price.

Nvidia has left four commercial terms unresolved:

• No effective date.
• No confirmed standard price.
• No defined software bundle.
• No disclosed number of GPUs under paid licenses.

The company also has not explained how the proposed fee would fit its existing enterprise software business.

How large could software revenue become?

FourWeekMBA outlined a scenario involving five million paid licenses. At five million paid licenses, annual revenue would reach $10 billion and five-year revenue would reach $50 billion, assuming every subscriber paid $2,000 and renewed without interruption.

Five million is a scenario input, not a subscriber forecast, and Nvidia has disclosed no paid-license base supporting it. Neither figure is presented as Nvidia guidance or contracted revenue.

Does this undercut Nvidia’s current software price?

Nvidia lists its AI Enterprise platform at $4,500 per GPU for a one-year subscription. That is more than twice Huang’s proposed $2,000 annual figure, which is less than half the listed AI Enterprise price.

The two prices share a one-year duration and per-GPU basis, but Nvidia has not established that they cover the same software products or licensing terms, so the comparison does not prove a discount. The lower number cannot be treated as a confirmed price cut.

What does OpenShell add to agent security?

Nvidia is developing infrastructure to deploy and manage AI agents in enterprise environments. Those systems can use tools, access data, and execute tasks, creating security and management challenges at business scale.

On October 21, Nvidia announced the OpenShell Reference Architecture. It combines Vera Rubin computing infrastructure, Vera CPUs, and BlueField-4 data processing units running Nvidia Sentry.

Nvidia says OpenShell provides a secure runtime for AI agents and establishes boundaries around their actions and access to tools and resources. It also says Sentry monitors agent activity independently and can quarantine an agent that attempts to cross those boundaries. According to Nvidia, the architecture places controls in the underlying infrastructure at runtime rather than depending solely on safeguards built into an AI model or application when agents access tools.

Nvidia had previously introduced its Open Agent Safety Platform in September. The newer reference architecture extends that work to enterprise infrastructure and edge systems.

Who is building around the controls?

Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents. The combination gives businesses more control over what AI agents can access inside execution sandboxes.

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

The arrangements add controls around third-party agent activity, but their commercial relationship to Nvidia’s proposed fee remains undefined. The announcements do not identify which OpenShell software would fall within Nvidia’s proposed annual per-GPU license, or how the license would relate to Claude Managed Agents and Slack Admin Bridge.

What does this mean for Nvidia’s strategy?

The proposed fee fits Nvidia’s push to build an AI infrastructure business spanning chips, software, and AI-agent security, extending a business beyond GPUs 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.”

Across healthcare, automotive, telecommunications, and life sciences, Nvidia says it supplies broader technology stacks rather than chips alone, giving the proposed software fee strategic relevance beyond accelerator sales.

Still, the company has not disclosed an adoption target or customer commitment for the fee. The commercial test is whether Nvidia discloses an effective date, defines the bundle and explains how $2,000 relates to the $4,500 AI Enterprise subscription. Without those terms, the $50 billion case remains a scenario rather than a forecast.