Nvidia thinks about a $2,000 fee for each GPU each year
Nvidia thinks about a fee of about $2,000 each year for each GPU for its agent software, but gave no products or date.
Nvidia CEO Jensen Huang told of the fee on October 22 at GTC Berlin. The fee is about $2,000 each year for each GPU (a processor for AI work). It is for all Nvidia software that operates AI agents, which use tools and do tasks with small help from persons. Nvidia wants software licenses to be a big part of its business.
What is new
Huang gave the price as “about $2,000 per GPU per year” for the software that operates agents on AI factories (large groups of Nvidia processors). On October 21, Nvidia showed OpenShell Reference Architecture, a safe area for agents to operate. It has a monitor, Sentry, that works by itself and can isolate an agent that goes through a safety limit.
The numbers
The source gives one sum. It is not a forecast by Nvidia. 5 million licenses that persons pay for, at $2,000, give $10 billion each year. The price of AI Enterprise is $4,500 for each GPU for one year. That is more than 2 times the new fee.
What is not known
Nvidia has not told which products the fee is for, when it starts, or how many GPUs can have it. It is not sure that $2,000 is the last price.
Insight: Nvidia may charge for the layer that operates its open models
Nvidia gave Flora, an open set of models, and Trend AI operates the open Nemotron model. Nvidia thinks about a fee for software that operates agents on its GPUs. This reads as a toll gate: open models get users, and the agent software seems to be where Nvidia collects. For leaders: count the software license in the cost of agent hardware.
Sources
Posted
