Overview
- NVIDIA CEO Jensen Huang said at GTC Berlin on Thursday that running the company’s software that powers AI agents “works out to about $2,000 per GPU per year.”
- NVIDIA has not confirmed whether the $2,000 figure maps to its existing AI Enterprise product, a narrower agent bundle, or a future licensing plan, and it has not announced any formal price change or timing.
- The cited number is roughly half of NVIDIA’s publicly listed AI Enterprise rate of about $4,500 per GPU for a one‑year subscription with support, creating immediate questions about scope and overlap.
- Reporters and analysts have run simple scenarios from the figure, noting that five million paid GPU licenses at $2,000 each would imply about $10 billion a year in recurring revenue and could scale much higher over multiple years under optimistic assumptions.
- Huang’s remark followed NVIDIA’s OpenShell and agent‑security discussion, which included Vera Rubin systems, BlueField‑4 DPUs and NVIDIA Sentry, and the next key developments to watch are formal pricing details, product definitions and how customers will be billed.