Nvidia Targets AI Software Subscription Revenue, Huang Renxun Proposes About $2,000 per GPU per Year

Nvidia Targets AI Software Subscription Revenue, Huang Renxun Proposes About $2,000 per GPU per Year

IT Home Download the Client | Exclusive First Look

October 26, 2026, 09:48:47 | From Shandong

Updated October 26, 2026, 10:36:01

IT Home, October 26 — Nvidia CEO Jensen Huang said the annual license fee for accompanying AI software is about $2,000 per GPU (IT Home note: at the current exchange rate, about 13,355 yuan), used to "run the full set of software Nvidia provides that supports AI agents operating on these AI factories." Video of the relevant remarks recently surfaced. According to a calculation scenario given by FourWeekMBA, if paid licenses reach 5 million, this pricing model is expected to generate $50 billion (at the current exchange rate, about 333.872 billion yuan) in revenue within five years.

Huang discussed the pricing on October 22 at the GTC conference in Berlin, where Nvidia also announced a more ambitious plan: to build an AI infrastructure business covering chips, software, and AI agent security capabilities. "Software licensing will gradually become a very important part of our business. Our licensing model is roughly like this... For example, to run Nvidia's full set of software for deploying agents on AI factories, the cost is about $2,000 per GPU per year (at the current exchange rate, about 13,355 yuan)," Huang said.

The scale of this revenue is easy to estimate. If 5 million GPUs are included in the paid licensing scope, at $2,000 per GPU per year (at the current exchange rate, about 13,355 yuan), Nvidia would take in $10 billion (at the current exchange rate, about 66.774 billion yuan) annually; as long as customers keep renewing, the five-year total could reach $50 billion (at the current exchange rate, about 333.872 billion yuan). But there are uncertainties here: Nvidia did not specify which GPUs the $2,000 (at the current exchange rate, about 13,355 yuan) quote applies to, nor did it confirm whether it is a uniform standard price. It is also unclear exactly which software this licensing covers and whether permissions overlap between different products.

Clarifying this point is crucial, because Nvidia already has paid enterprise software products. Its AI Enterprise platform is listed at an annual subscription price of $4,500 per GPU (at the current exchange rate, about 30,049 yuan). Although the $2,000 (at the current exchange rate, about 13,355 yuan) mentioned by Huang is lower, Nvidia has not confirmed that the two prices correspond to exactly the same products and licensing terms, so it cannot be directly treated as an official price cut.

The context for this discussion of software licensing is that Nvidia is developing more supporting infrastructure to make it easier for enterprises to deploy and manage AI agents. Such systems can call tools, read data, and automatically execute tasks. When enterprises roll them out at scale, new security and management challenges will arise.

On October 21, Nvidia released the OpenShell reference architecture, a system that integrates Vera Rubin computing infrastructure, Vera CPU, and the BlueField-4 data processing unit (DPU) equipped with Nvidia Sentry security components. OpenShell provides a secure operating environment for AI agents, defining the boundaries of agent actions and restricting their permissions to call tools and access resources. Nvidia says Sentry can independently monitor agent behavior and isolate an agent once it attempts to cross the boundary. This architecture pushes security controls down to the underlying infrastructure rather than relying solely on protections inside AI models or applications.

Nvidia launched an open agent security platform as early as September, and this new reference architecture extends from the former, expanding the control approach to enterprise infrastructure and edge devices.

Other vendors are also adding similar control capabilities to their products. Anthropic combines OpenShell with BlueField hardware and Claude managed agents, making it easier for enterprises to control what AI agents can access within an execution sandbox. Salesforce has developed a Slack management bridge tool adapted for OpenShell, allowing administrators to review agent access requests, approve or deny permissions, and monitor runtime activity directly in Slack.

This new software pricing also echoes Huang's long-standing view: Nvidia is no longer just a GPU supplier. The company's business has expanded to networking equipment, complete systems, enterprise software, and infrastructure for AI workloads. Huang recently said on the BG2Pod podcast: "Someday, people may no longer classify us as a chip company." He cited examples of Nvidia having entered healthcare, automotive, telecommunications, life sciences, and other industries, providing a complete technology stack rather than just chips.

Even so, Nvidia has not yet announced when the $2,000 per GPU per year (at the current exchange rate, about 13,355 yuan) charging plan will take effect, which software it includes, or how this pricing will connect with its existing enterprise software business.

"Special statement: The above content (including videos, pictures, or audio if any) is uploaded and published by users of 'Dafeng Hao,' a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."