Nvidia has made billions from supplying the GPUs powering the AI boom, but hardware may only be one part of its long-term strategy. The company is increasingly focusing on software that runs on top of its chips, and CEO Jensen Huang has now discussed a potential annual licensing model that could create another major source of recurring revenue.

Speaking at GTC Berlin on October 22, Huang said software licensing is expected to become an increasingly important part of Nvidia’s business. As an example, he suggested a cost of around $2,000 per GPU per year for Nvidia software used to run AI agents inside what the company calls “AI factories,” essentially large data centers built to handle AI workloads.

A new expense for AI firms?

For companies operating thousands of Nvidia GPUs, that could quickly turn into a significant annual expense. A server farm with 10,000 licensed GPUs, for instance, would be looking at around $20 million per year at the $2,000 figure. At one million GPUs, that would translate to $2 billion in annual licensing revenue for Nvidia.

However, the $2,000 figure should not be treated as a confirmed subscription price just yet. Nvidia hasn’t announced a product with that exact pricing, nor has it provided a launch date or explained precisely which software would be included. Huang’s comments instead give an indication of how Nvidia could price its growing software ecosystem as AI agents become more widely deployed.

Nvidia already charges businesses for enterprise software. Its existing Nvidia AI Enterprise offering is listed at $4,500 per GPU for a one-year subscription, while a five-year subscription costs $18,000 per GPU. That makes Huang’s $2,000 example noticeably cheaper, but Nvidia hasn’t clarified whether the proposed licensing model would replace AI Enterprise, work alongside it, or cover an entirely different collection of tools.

The company has also been expanding the software and security infrastructure surrounding AI agents. It recently introduced its OpenShell Reference Architecture, designed to help companies control what autonomous AI agents can access and do.

For Nvidia, an annual software fee could also provide something that hardware sales don’t: predictable recurring revenue. GPUs used in AI server farms can cost tens of thousands of dollars, but those are primarily hardware purchases. Charging customers every year for the software running on those GPUs would allow Nvidia to continue generating revenue from the same hardware long after it has been installed.

About the Author

Abdul Haddi

Abdul is a tech writer and Editor for PC Guide, specializing in all things tech, gaming, and hardware.