Nvidia floats $2,000-per-GPU annual fee for AI-agent software

Nvidia CEO Jensen Huang suggested at GTC Berlin that running Nvidia’s AI-agent software could cost about $2,000 per GPU per year, a figure he offered as a rough example rather than a formal price announcement. The comment ties directly to Nvidia’s push around agentic AI and the company’s OpenShell reference architecture.

Huang floated the $2,000-per-GPU number while outlining how Nvidia wants software licensing to become a larger share of its business, but Nvidia has not published a product price or an effective date for such a fee. Reporting from TweakTown and DiarioBitcoin says the CEO presented the figure as an estimate and did not specify which software packages it would cover.

An empty server rack aisle with anonymous rack doors open and rows of unbranded GPU cards mounted inside, cool white LED indicators glowing; conveys data-center scale and infrastructure focus

The $2,000 example sits below the company’s publicly listed AI Enterprise subscription price, which Nvidia documents at $4,500 per GPU for a one-year license. Both TweakTown and DiarioBitcoin note that the $2,000 number doesn’t match Nvidia’s current published prices and may refer to a narrower agent-focused bundle rather than the full AI Enterprise suite.

Huang introduced the remark alongside Nvidia’s OpenShell reference architecture and tools such as Nvidia Sentry, which the company describes as mechanisms to contain and monitor autonomous agents. DiarioBitcoin reports OpenShell is designed to limit agents’ access to resources, while Sentry can quarantine agents that attempt unauthorized actions.

Close view of an anonymous technician’s hands adjusting an unbranded network cable in front of blurred rack-mounted DPUs and GPU modules, no logos visible; suggests infrastructure work to secure agentic AI

Industry coverage frames the $2,000 figure as illustrative: outlets say it highlights Nvidia’s strategy to monetize software and security for agentic AI rather than a confirmed licensing policy. Nvidia has not issued a separate public pricing announcement tying OpenShell or Sentry to that annual per-GPU fee.

For readers tracking Nvidia’s wider AI strategy and investor implications, this shift toward recurring software fees was part of Huang’s GTC Berlin keynote; our site previously covered his appearance at GTC Berlin and the company’s broader AI positioning. Jensen Huang delivers GTC 2026 keynote in Berlin and a recent piece on investors watches for AI demand provide related context for NVDA watchers. NVDA stock: what investors are watching as AI demand reshapes Nvidia’s outlook

Sources

  • TweakTown — reported Huang’s $2,000-per-GPU estimate from his GTC Berlin keynote and explained it appears to be illustrative, noting Nvidia’s AI Enterprise list price.
  • DiarioBitcoin — reported the same GTC Berlin remark, cited OpenShell and Sentry as the agent-control architecture, and noted Nvidia did not confirm scope or timing for the fee.

Give your feedback

★★★★★

Be the first to rate this post
or leave a detailed review



ECIKS.org is an independent media. Support us by adding us to your Google News favorites:

Post a comment

Publish a comment