$2,000 Per GPU: Nvidia Weighs New AI Agent Software Fee

Nvidia has discussed charging data-center customers $2,000 per GPU each year for a new software product described as running "agents on AI factories," according to Tom's Hardware. The outlet reports that the figure comes from internal discussions and that Nvidia has not finalized pricing or scope.

What software would the $2,000 fee cover?

Tom's Hardware describes the product only by the phrase Nvidia has reportedly used internally.

The fee would apply to software that "runs agents on AI factories," according to Tom's Hardware's reporting on the discussions.

The outlet notes that details beyond that description are unclear, including what specific functions the software performs and which customers would be required to pay for it.

How does $2,000 compare to Nvidia's existing AI Enterprise fee?

Nvidia already sells a software platform called AI Enterprise, and Tom's Hardware reports the company charges more than twice the proposed $2,000-per-GPU figure for that existing product on a per-GPU, per-year basis. That gap is central to the story: a new $2,000 charge would sit well below what Nvidia already collects for AI Enterprise, raising the question of why a buyer would need both.

What's unclear about overlap between the two products?

Tom's Hardware's reporting flags the overlap issue directly, noting that it is not clear whether the agent-running software would duplicate capabilities already included in AI Enterprise or whether it represents a genuinely separate layer of the stack. Nvidia has not published a feature comparison between the two, and the outlet did not report an on-record statement from Nvidia addressing the distinction.

Why would Nvidia add another per-GPU software charge?

Nvidia's GPU hardware revenue has for years come bundled with an expanding set of software offerings, from CUDA libraries to AI Enterprise, that the company sells separately from the chips themselves. A new, lower-priced tier aimed specifically at running AI agents would give Nvidia a second recurring revenue line tied to GPU deployments, on top of hardware sales and the existing AI Enterprise subscription. Tom's Hardware's framing suggests the move is still at the discussion stage rather than a settled product launch, meaning pricing, packaging, and the AI Enterprise overlap could change before any public rollout.

What should buyers of Nvidia GPUs watch for next?

Enterprise customers planning large GPU deployments have a direct financial interest in how Nvidia resolves the overlap question, since a $2,000-per-GPU charge stacked on top of AI Enterprise pricing would materially change the total cost of running agent-based AI workloads at scale. Tom's Hardware's report does not specify a timeline for an announcement, and Nvidia has not issued its own public pricing sheet for the discussed product. Until the company clarifies what the new fee covers and whether it replaces or supplements AI Enterprise, buyers evaluating multi-year GPU contracts have only the broad outline reported so far to work from.

The original Tom's Hardware report, carried via Google News, is the sole source for the pricing figures discussed here; Nvidia has not issued a separate public statement confirming the details.