Nvidia CEO Jensen Huang has outlined a potential annual software licensing fee of roughly $2,000 per GPU for tools that run AI agents on its hardware, signaling a strategic shift toward recurring software revenue. Huang discussed the pricing during a Q&A session at GTC Berlin on October 22, 2026.
"Software licensing will become an increasingly major part of our company's business. For example, it's about $2,000 per GPU per year to run all the software that Nvidia provides to power agents on these AI factories," Huang stated.
The fee is tied to Nvidia's agent-running software stack, which centers on the OpenShell reference architecture - a secure runtime that sandboxes AI agents - and Nvidia Sentry, which runs on BlueField-4 DPUs to monitor activity and quarantine agents that breach security boundaries.
The infrastructure is supported by Vera Rubin systems and Vera CPUs. The revenue implications are significant; at that rate, 5 million paid licenses would generate $10 billion annually.
However, Nvidia has not confirmed if this fee is a final standard or how it interacts with the existing AI Enterprise platform, which costs $4,500 per GPU annually. Consequently, the $2,000 figure should not be interpreted as a price reduction.