Nvidia is exploring a Nvidia AI agents cost of about $2,000 per graphics processing unit (GPU—the specialised chip that powers much of today’s AI) each year for software that runs AI agents on its hardware. Chief executive Jensen Huang described the idea at GTC Berlin on 22 October as the company looks to expand its software business alongside chip sales.
Huang said software licensing would become a major part of Nvidia’s business. In his words, the company might licence “about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories.” Nvidia has not said which products would carry that price, or when any such fee would begin.
How the proposed fee sits next to existing software
Nvidia already sells its AI Enterprise software platform for $4,500 per GPU for a one-year subscription. That list price is more than double the $2,000 figure Huang mentioned for agent-related software. It is not yet clear whether the two prices would cover different products, different licence terms, or overlapping stacks. Until Nvidia maps the new proposal onto its catalogue, buyers cannot treat the $2,000 number as a firm replacement for AI Enterprise.
The Nvidia AI agents cost discussion also came with a large illustrative revenue example, not a company forecast: if five million licences were sold at $2,000 each per year, that would equal $10 billion a year, or $50 billion over five years if customers kept renewing. Nvidia has not said how many GPUs might qualify or whether $2,000 will be the final price.
Why AI agents matter for the software push
AI agents are systems that can use tools, reach data, and complete tasks with little human oversight. Firms that run them at scale need software to schedule workloads and to stop agents from touching systems they should not reach. That demand is the commercial backdrop Huang pointed to when he framed licensing as a growing share of Nvidia’s business.
Timeline: what has happened so far
- 21 October: Nvidia introduced its OpenShell Reference Architecture, aimed at giving AI agents a more secure setting and limiting which tools and resources they can use. The design uses Nvidia’s Vera Rubin computing system, Vera CPUs, and BlueField-4 data processing units (DPUs) with Nvidia Sentry, which can monitor agent activity and isolate agents that cross set security limits.
- 22 October: At GTC Berlin, Jensen Huang outlined a possible Nvidia AI agents cost of about $2,000 per GPU per year for software that runs agents on Nvidia’s AI systems, while stressing software licensing as a larger future business line.
What is still unknown
Key details remain open: which software packages would fall under the $2,000-per-GPU annual model, how it would relate to the $4,500 AI Enterprise subscription, how many GPUs would need a licence, and when customers would first be billed. Until those answers arrive, the figure is a directional signal from the chief executive, not a finished price list.
Aasan alfaaz mein
Nvidia ke CEO ne kaha hai ke AI agents chalane wale software ke liye har GPU par salana qareeb $2,000 fee ho sakti hai. Abhi yeh final rate nahi, aur na yeh bataya gaya ke kab shuru hogi. Pehle se AI Enterprise software $4,500 per GPU salana milta hai. Pakistan mein jo companies Nvidia chips par AI chalati hain, unhen baad mein software licence ki alag cost dekhni pad sakti hai.
