Nvidia chief executive Jensen Huang has described the company’s agent software as costing about $2,000 per GPU per year, according to a Tom’s Hardware report published on October 25. Huang made the remark at GTC Berlin, and Nvidia has not confirmed the price, which products it would cover or when it would start, so we treat the figure as a reported discussion rather than an announced price list. It matters because the number would turn Nvidia’s installed base of GPUs into a recurring software revenue stream.

What Huang said

Tom’s Hardware, citing a video of the comments, quotes Huang saying that software licensing “will become increasingly a major part” of the company’s business. In the same passage he put the cost at “about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories.” The outlet places the remarks at GTC Berlin on October 22.

The report adds several caveats. Nvidia has not said how many GPUs the fee would apply to, whether $2,000 is a standard price, which software would be included or whether products would overlap. Tom’s Hardware also notes that a headline-friendly projection circulating online, $10 billion a year if five million GPUs carried paid licenses, comes from a third-party scenario and not from Nvidia. We could not watch the video, so the wording relies on Tom’s Hardware.

How it compares with NVIDIA AI Enterprise

Nvidia already sells a paid software suite, NVIDIA AI Enterprise. The product page itself shows no prices and points visitors to a sales contact and a 90-day trial license. NVIDIA’s separate licensing guide, however, lists self-managed prices that apply per GPU:

  • One year: $4,500 list price per GPU, or $1,125 for qualifying education and Inception or Connect program members.
  • Two and three years: $9,000 and $13,500 per GPU.
  • Five years: priced at the four-year rate of $18,000.
  • Perpetual license with five years of support: $22,500 per GPU.
  • Cloud marketplaces: $1 per GPU per hour, plus the cloud provider’s instance cost.
An aisle between metal equipment racks in a server room
A generic server-room aisle: per-GPU licenses would be counted across racks like these, whatever the vendor. This is an illustrative photo, not an Nvidia facility. Photo: Ismail Enes Ayhan / Unsplash

On those list prices, $2,000 per year is less than half of the one-year AI Enterprise rate. That does not mean a price cut. Nothing in the report says the new figure covers the same software or the same support terms, and the guide’s prices are described as list prices that can differ by channel. Until Nvidia publishes a product definition, the two numbers are not directly comparable.

The software it may refer to

Huang’s phrase, software that “runs agents,” lines up with Nvidia’s recent agent security push. On October 21 the company announced an OpenShell Reference Architecture, which combines Vera Rubin systems, Vera CPUs for agent harnesses, the open-source OpenShell runtime and BlueField DPUs running a framework called Sentry. Nvidia says Sentry watches agent behavior from outside the agent’s own environment and can quarantine an agent that crosses its boundaries within milliseconds, and it claims up to 20% more agent capacity. The release says the architecture is available now and does not mention pricing. Our earlier story on the NVIDIA Agent Toolkit and Confidential Bare Metal announcements at GTC Berlin covers the surrounding software stack.

Whether the $2,000 figure refers to OpenShell, Sentry, the Agent Toolkit or a bundle is not stated.

Why buyers and investors care

For enterprises, a per-GPU subscription adds a fixed yearly line on top of hardware and power. Finance teams already struggle to forecast agent costs, a theme we explored in how finance teams plan enterprise AI spend. A flat per-GPU fee is easier to model than token billing, but it is owed whether or not a GPU is busy.

For investors, the appeal is recurring revenue that grows with every GPU shipped. The scale remains speculation until Nvidia discloses attach rates, since only customers running agents on their own infrastructure would plausibly pay.

What to watch next

  • Whether Nvidia publishes a price page or licensing guide entry for the agent software, and which products it names.
  • If the fee replaces, bundles with or sits beside NVIDIA AI Enterprise at $4,500 per GPU per year.
  • How Nvidia describes the paid portion of open-source components such as OpenShell.
  • Comments on software revenue in Nvidia’s next earnings call.

Key takeaways

  • Huang reportedly put Nvidia’s agent software license at about $2,000 per GPU per year at GTC Berlin.
  • Nvidia has not confirmed pricing, scope or start date, so the figure remains unconfirmed.
  • NVIDIA’s licensing guide lists AI Enterprise at $4,500 per GPU for a one-year subscription.
  • The two prices cannot be compared directly until Nvidia says what each covers.
  • The revenue projections circulating online are third-party scenarios, not Nvidia guidance.

Frequently asked questions

How much would Nvidia’s agent software cost per GPU?

Huang reportedly said about $2,000 per GPU per year. Nvidia has not published this as a price, and it has not said which software is included.

How does that compare with NVIDIA AI Enterprise?

NVIDIA’s licensing guide lists a one-year AI Enterprise subscription at $4,500 per GPU. The reported $2,000 is lower, but it is unclear whether the two cover the same products.

Has Nvidia confirmed the fee?

No. The figure comes from Huang’s remarks as reported by Tom’s Hardware, and the company has not announced a launch date or product details.