NVIDIA targets recurring software revenue with new GPU licensing estimates

A shift toward software revenue

NVIDIA CEO Jensen Huang recently shared a new perspective on how the company plans to monetize its AI ecosystem. During a Q&A session at GTC Berlin, Huang discussed the growing importance of software licensing as a core part of the company's business strategy. He specifically highlighted the software required to power AI agents in what he calls AI factories.

Huang suggested that the software side of the business could be valued at approximately $2,000 per GPU per year. While this was presented as a rough example during the talk, it provides a clear window into NVIDIA's long-term financial goals. The company wants to move beyond just selling the physical chips that train and run models.

By establishing a recurring revenue model for software, NVIDIA aims to create a more predictable income stream. This strategy complements their current dominance in the hardware market. It ensures that as companies deploy more AI infrastructure, NVIDIA captures a portion of the ongoing operational costs as well.

Comparing the new figures

The $2,000 figure mentioned by Huang has caused some confusion because it does not align with NVIDIA's current official pricing. According to the company's own licensing guide, the AI Enterprise suite is currently priced much higher. Specifically, the guide lists a one-year subscription with support at $4,500 per GPU.

There is also a five-year subscription option available that costs $18,000 per GPU. Because the $2,000 figure is lower than the current retail price, it is unclear if this represents a price cut. It might instead refer to a different, narrower bundle specifically for AI agents or a future licensing model that has not been finalized.

As TechPowerUp reported, there is currently no official word on the timing of these changes. Because the numbers are inconsistent with existing products, the figure should not be viewed as an immediate price drop. It serves more as a benchmark for the value NVIDIA expects to extract from its software stack.

The OpenShell architecture

Alongside these revenue discussions, NVIDIA announced a new technical framework called the OpenShell Reference Architecture. This system is designed to work with Vera Rubin systems for inference and Vera CPUs for running agents. It also incorporates BlueField-4 DPUs to run NVIDIA Sentry, which monitors the actions of those agents.

OpenShell is an open-source runtime that provides a sandbox for AI agents. A sandbox is a secure, isolated environment where code can run without affecting the rest of the system. This is a critical component for companies that want to deploy autonomous agents while maintaining control over their internal data and security.

Huang pointed to OpenShell specifically when asked about the issue of AI safety. By providing a standardized way to run and monitor agents, NVIDIA is positioning its software as the necessary foundation for safe deployment. This architecture helps bridge the gap between raw hardware power and usable, secure AI applications.

The scale of the opportunity

The financial implications of NVIDIA's software strategy are significant when looking at the scale of the current AI boom. If NVIDIA successfully licenses its software at the $2,000 per GPU rate, the math becomes very clear for the company's bottom line. Every million GPUs licensed at this rate would generate $2 billion in annual revenue.

This recurring revenue is particularly attractive because it creates a

Clarifying the licensing ambiguity

The specific nature of the $2,000 figure remains a point of uncertainty for industry observers. Because NVIDIA already offers the AI Enterprise suite on a per-GPU basis, the new number creates a confusing overlap with existing products. It is not currently clear if this amount refers to the same software suite or a different offering.

NVIDIA has not specified whether this figure represents a narrower bundle designed specifically for AI agents or a completely new licensing model for the future. Because of this lack of detail, the $2,000 amount should not be interpreted as a price cut on current services. The company has yet to provide a timeline for when these specific software costs might come into effect for customers.