Speaking at a Q&A session at GTC Berlin on October 22, 2026, Nvidia CEO Jensen Huang said that running the full stack of Nvidia software behind AI agents in large data centers costs about $2,000 per GPU per year. He framed the figure as an illustrative example of how the company intends to price its software, not as an announced product or price list.
The number matters for planning purposes because Nvidia has not yet clarified which software titles the $2,000 would cover, when any such fee would take effect, or how it would relate to the existing Nvidia AI Enterprise suite, which already lists at $4,500 per GPU for a one-year subscription. The relationship between the two prices has not been explained, and no launch date has been given. Nvidia has not published the $2,000 figure in a press release, earnings call, or formal product page.
For organizations building or expanding AI infrastructure on Nvidia hardware, this is a signal worth tracking as a potential future budget line. A deployment of 1,000 GPUs, for example, would represent $2 million in annual software costs at that rate, separate from hardware and cloud compute charges. Businesses in that position may want to ask their Nvidia account team or reseller how any forthcoming software licensing would interact with their current AI Enterprise agreements.
Sources
Does this affect your business?
Our advisors can help you figure out the right move, at no cost.