
ⓘ Nvidia
Nvidia is considering a $2,000 annual software fee per GPU for running AI agents. The proposed licensing model could generate billions in recurring revenue, adding to the company's existing AI software business, although pricing and rollout details remain unclear.
Nvidia is considering charging $2,000 per GPU annually for software used to run AI agents on its chips. CEO Jensen Huang discussed the potential fee at GTC Berlin on October 22, as the company looks to make software a bigger part of its business. Nvidia has yet to clarify which products the proposed pricing would cover or when it might take effect.
Software licensing will become increasingly a major part of our company’s business. The way that we license will be…for example, it’s about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories, - Jensen Huang.
The potential revenue is significant. At $2,000 per year, five million paid licenses would bring in $10 billion annually, or $50 billion over five years if customers continued renewing their subscriptions. These figures are illustrative calculations, not Nvidia's revenue forecast: the company has not disclosed how many GPUs would be eligible for the proposed license or whether $2,000 would be its standard or final price.
Nvidia already sells its AI Enterprise software platform, which costs $4,500 per GPU for a one-year subscription. That's more than twice the proposed fee, although the two prices may cover different software and licensing terms. It's therefore unclear how the new proposal would fit into Nvidia's existing offerings.
The company is looking to capitalize on growing interest in AI agents, which can use tools, access data, and carry out tasks with limited human intervention. Businesses deploying these systems at scale need software to manage their workloads and prevent agents from accessing resources they shouldn’t.
On October 21, Nvidia announced its OpenShell Reference Architecture, designed to provide a secure runtime for AI agents and restrict their access to tools and resources. The architecture combines Nvidia’s Vera Rubin computing infrastructure, Vera CPUs, and BlueField-4 data processing units (DPUs) running Nvidia Sentry. The latter monitors agent activity independently and can quarantine agents that attempt to cross established security boundaries.
Andrew Sozinow - Tech Writer - 164 articles published on Notebookcheck since 2024
I’ve been fascinated by computers, electronics and modern technology since childhood. I started writing IT-news at high school and have been doing it continuously for more than 10 years. During this time, I have worked for many media outlets, and now I am a news editor at 3DNews. Sometimes I also write smartphone reviews. In July 2024, I decided to try my hand at writing news on Notebookcheck. When I'm not working, I like to play videogames, do puzzles, and travel.
Andrew Sozinov, 2026-10-25 (Update: 2026-10-25)

