Nvidia Floats $2,000 per GPU Annually for Its AI Software
Jensen Huang spoke of a roughly $2,000 per-GPU annual license for the software stack that runs AI agents. What the fee would cover remains unclear.
Merhabyte Newsdesk · · 3 min read
Nvidia CEO Jensen Huang has spoken of a license fee of roughly $2,000 per GPU per year for the company's AI software. In Huang's telling, the fee covers the Nvidia-supplied software that runs agents on AI factories. He made the remarks on October 22 at the GTC Berlin event.
Huang also stressed that software licensing will account for a growing share of Nvidia's business. But Nvidia has given no detail on whether the figure is a standard price, which software it would cover, or when it would take effect.
What the numbers could mean
A simple calculation shows what is at stake: $2,000 per GPU per year, applied to five million licensed GPUs, would mean $10 billion in annual revenue for Nvidia. If customers kept renewing their subscriptions, that would reach $50 billion over five years. The five-million-license scenario is a hypothetical that circulated after the event, not a target Nvidia has announced.
One assessment shared on social media assumed that half of the 10 million enterprise GPUs shipped each year would take paid software, and arrived at the same $10 billion in potential recurring annual revenue. By that reasoning, every new year of GPU sales could also create a separate software revenue stream. Nvidia, however, has not said how many GPUs the $2,000 fee might apply to.
Relationship to existing enterprise pricing is unclear
Nvidia already sells enterprise software. Its AI Enterprise platform is listed from $4,500 per GPU for a one-year subscription. The $2,000 figure Huang mentioned looks lower, but Nvidia has not confirmed that the two prices cover the same products or the same license terms. The number should therefore not be read as a confirmed price cut.
It is also unclear which software the license would include and whether products would overlap. That distinction matters both for the company's revenue model and for customers' cost calculations.
OpenShell and the focus on agent security
The licensing discussion comes as Nvidia expands the infrastructure needed to deploy and manage AI agents in enterprise environments. These systems can use tools, access data and carry out tasks, which creates new security and governance problems when they run at scale. On October 21, Nvidia announced the OpenShell Reference Architecture, which combines Vera Rubin compute infrastructure, Vera CPUs and BlueField-4 data processing units (DPUs) running Nvidia Sentry.
According to Nvidia, OpenShell provides a secure runtime for agents, setting boundaries on their actions and on their access to tools and resources. Sentry independently monitors agent activity and can quarantine an agent that tries to cross those boundaries. The aim is to move security controls into the infrastructure itself rather than leaving them solely to safeguards in the AI model or application. Nvidia introduced its Open Agent Safety Platform in September; the new architecture builds on that work and extends the approach to enterprise infrastructure and edge systems.
Other companies are building these controls into their products. Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents to give businesses additional control over what agents can access in their execution environments. Salesforce has developed a Slack Admin Bridge that lets administrators review and approve or reject agent access requests directly in Slack and monitor activity.
Is Nvidia's identity as a chip company changing?
The software pricing Huang described fits the broader thesis that Nvidia is moving beyond being a GPU supplier. The company has expanded into networking, systems, enterprise software and infrastructure designed for AI workloads. Huang said recently on the BG2Pod podcast that people may one day stop classifying Nvidia as a chip company, noting that in sectors such as healthcare, automotive, telecommunications and life sciences the company offers broader technology stacks rather than just chips.
Even so, Nvidia has not said when the $2,000 per-GPU fee would take effect or which software it would cover. How the fee Huang described would fit into the existing enterprise software business is also unknown for now.
