Nvidia’s $2,000-Per-GPU Software Fee Puts AI Agent Infrastructure in Focus

| 2 Min Read
Nvidia is weighing annual software pricing of about $2,000 per GPU as it builds infrastructure to secure and manage enterprise AI agents.

Nvidia’s roughly $2,000-per-GPU annual software figure points to a larger shift: the company wants to sell the systems that run enterprise AI agents, along with the chips beneath them. The commercial scope remains unclear, but the price highlights how software could become a bigger part of Nvidia’s business.

CEO Jensen Huang discussed the figure at GTC Berlin on October 22, describing a fee for “all the software that Nvidia provides that runs agents on these AI factories.” He said software licensing would become a more significant part of the company’s business.

The price is a signal, not a confirmed package

At $2,000 per GPU each year, five million paid licenses would amount to $10 billion in annual revenue, or $50 billion over five years if customers renewed throughout that period. That scenario, outlined by FourWeekMBA, depends on assumptions rather than disclosed Nvidia sales or licensing plans.

Nvidia hasn’t said how many GPUs might qualify, whether $2,000 is a standard rate, or which products the fee would include. Those details matter because the company already lists its AI Enterprise platform at $4,500 per GPU for a one-year subscription. Huang’s figure is lower, but Nvidia hasn’t established that the two prices cover the same software or terms. It shouldn’t be read as a confirmed discount.

Security moves into the infrastructure

The proposed fee arrives as Nvidia adds tools for deploying and governing agents, which can access company data, use software tools and carry out tasks. At scale, those capabilities create operational and security demands beyond the model itself.

On October 21, Nvidia announced its OpenShell Reference Architecture. It combines Vera Rubin computing infrastructure, Vera CPUs and BlueField-4 data processing units running Nvidia Sentry. OpenShell sets boundaries for agent actions and access to tools and resources. Nvidia says Sentry monitors activity independently and can quarantine an agent that tries to cross those limits. The design places controls in the infrastructure, alongside safeguards in models and applications.

The architecture extends Nvidia’s Open Agent Safety Platform, introduced in September, to enterprise infrastructure and edge systems. Other vendors are connecting these controls to their own products. Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents to give businesses additional control over agent access inside execution sandboxes. Salesforce has built a Slack Admin Bridge for OpenShell, where administrators can review permission requests, approve or deny them, and monitor agent activity.

Nvidia’s business reaches beyond GPUs

Huang’s pricing comments fit Nvidia’s push into networking, systems, enterprise software and AI infrastructure. On the BG2Pod podcast, he said, “One of these days, people might not categorize us as a chip company anymore.” He cited Nvidia’s work across healthcare, automotive, telecommunications and life sciences, where it sells broader technology stacks.

For enterprise buyers, the practical question is what the proposed fee includes and how it fits with existing licenses. Nvidia hasn’t announced a start date or clarified the software covered. Until it does, the $2,000 figure is best read as a marker of where the company sees value: managing and securing the AI systems built around its GPUs.

Source: Newton Gitonga · www.tomshardware.com

Comments

Please sign in to comment.