Nvidia Floats $2,000-Per-GPU Annual Fee for AI Agent Software

Nvidia CEO Jensen Huang put a number on the software layer that runs AI agents last week, and the infrastructure industry is still working out what, exactly, he meant.

Speaking during a Q&A at GTC Berlin on October 22, Huang said it costs about $2,000 per GPU per year to run Nvidia's full software stack for AI agents. As Tom's Hardware reported, the figure came without a product sheet. Huang framed it as an illustration, not a price list.

"Software licensing will become increasingly a major part of our company's business," he said, according to remarks cited by Tom's Hardware. "The way that we license will be -- for example, it's about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories."

The word "example" is doing a lot of work there. As Tom's Hardware noted, Nvidia hasn't said which programs the $2,000 covers, when it would apply, or how it would relate to the AI Enterprise suite the company already sells per GPU. That existing platform already lists at $4,500 per GPU for a one-year subscription, and $18,000 for five years, making Huang's lower figure confusing for anyone trying to budget against current contracts.

In Nvidia's framing, "AI factories" are data centers built to produce AI output at scale, the places where an agent -- a program that takes actions autonomously using tools and data -- runs across thousands of GPUs at once. A per-GPU annual license, if it lands, would convert Nvidia's installed hardware base into a subscription annuity. The math analysts are running: five million GPUs licensed at $2,000 each comes to $10 billion a year. That's illustrative arithmetic, not a forecast Nvidia has published, but it explains why the comment travelled.

The key thing missing here is the actual product. Nvidia hasn't launched anything with this price attached. There's no SKU, no effective date, no clarification of whether this fee would layer on top of AI Enterprise or replace part of it. One day before the GTC Berlin Q&A, Nvidia unveiled a broader software and hardware stack for running agents securely inside enterprises, which several outlets linked to the pricing signal -- but Nvidia itself hasn't confirmed the connection.

For enterprise buyers, the uncertainty is the real news. Infrastructure teams planning AI deployments are already working through hardware costs, model API fees, and the operational overhead of keeping agents secured and governed. A recurring per-GPU software charge of this size, if it becomes real, would be a material budget line, not a rounding error. As AI Weekly noted, founders and IT buyers should avoid assuming agent deployments will carry a linear cost curve -- licensing, isolation, monitoring, and governance are all candidates to become significant expense categories.

What Huang signaled in Berlin is that Nvidia wants to own that layer commercially, not just technically. Whether the $2,000 number survives contact with actual contract negotiations is a different question entirely. The company has given no timeline.

Sources cited:
- Tom's Hardware (https://www.tomshardware.com/pc-components/gpus/nvidia-discusses-usd2-000-per-gpu-annual-software-fee-for-running-agents-on-ai-infrastructure-firm-already-charges-more-than-twice-that-amount-for-its-ai-enterprise-software-platform)
- AI Weekly (https://aiweekly.co/alerts/huang-floats-2000-per-gpu-yearly-fee-for-nvidia-agent-software)
- Startup Fortune (https://startupfortune.com/nvidia-floats-charging-2000-a-year-per-gpu-for-its-ai-agent-software/)