NVIDIA discusses $2,000-per-GPU annual fee for software that 'runs agents on AI factories'
CEO Jensen Huang mentioned a potential $2,000 annual fee per GPU to run AI agent software on 'AI factories'.
At the GTC Berlin conference, NVIDIA CEO Jensen Huang discussed a potential shift in the company's revenue model. He suggested an annual licensing fee of approximately $2,000 per GPU to run the software required for operating AI agents on what he termed "AI factories."
Transitioning from Chips to AI Infrastructure
This move highlights NVIDIA's ambition to evolve beyond being just a hardware vendor. Huang emphasized that software licensing is set to become a major component of the company's business. The proposed model aims to support the entire software stack needed to manage AI agents—autonomous systems capable of using tools and accessing data to execute complex tasks.
The financial implications are significant. Industry analysts suggest that if NVIDIA manages to secure five million paid licenses at this rate, it could generate $10 billion in annual recurring revenue, potentially totaling $50 billion over five years. This strategy aligns with the company's broader push into networking, systems, and enterprise-grade AI infrastructure.
Security and the OpenShell Architecture
As AI agents become more integrated into enterprise workflows, security and governance become critical. To address these challenges, NVIDIA recently introduced the "OpenShell" Reference Architecture. This system integrates Vera Rubin computing infrastructure, Vera CPUs, and "BlueField-4" DPUs with Nvidia Sentry security monitoring.
Sentry is designed to monitor agent activity independently, providing the ability to quarantine any agent that attempts to exceed its authorized boundaries. This approach moves security controls directly into the underlying infrastructure, rather than relying solely on application-level safeguards. Companies like Anthropic and Salesforce are already exploring similar integrations to manage agent access within execution sandboxes.
Unclear Details and Existing Pricing
While the $2,000 figure is a significant talking point, several details remain unconfirmed. It is unclear whether this fee will overlap with existing products or if it represents a new tier of service. Notably, NVIDIA's current AI Enterprise platform is priced at approximately $4,500 per GPU for a one-year subscription. It remains to be seen whether the discussed $2,000 fee is a price reduction for certain services or a separate licensing model for the emerging AI agent market.
Гика
