Jensen Huang Says Nvidia Software for Running AI Agents in AI Factories Could Cost 2,000 Dollars Per GPU Per Year
NVIDIA may be taking the “you bought the GPU, now pay for the software” model to AI. Jensen Huang said NVIDIA’s software for running AI agents in its AI factories could cost around $2,000 per GPU per year. That would be on top of the cost of the GPUs themselves. At 1 million GPUs, that’s potentially $2 billion a year in software revenue. At 5 million, it would reach $10 billion per year. Huang also said software licensing will become an increasingly important part of NVIDIA’s business. NVIDIA already charges $4,500 per GPU/year for its AI Enterprise subscription
Context
Layering a 2,000 dollar annual per GPU software license for autonomous agents on top of hardware sales and existing Nvidia AI Enterprise subscriptions would turn deployed GPU fleets into multi billion dollar recurring software revenue streams.
Nvidia already charges enterprises 4,500 dollars per GPU per year for its Nvidia AI Enterprise software suite, which bundles NIM inference microservices, NeMo frameworks, and enterprise support.
Deep dive
Nvidia chief executive Jensen Huang stated that Nvidia's software stack for running autonomous AI agents inside large scale AI factories could be priced around 2,000 dollars per GPU per year on top of the upfront hardware purchase price, as highlighted on X by Pirat_Nation on Monday morning.
At a scale of 1 million deployed GPUs running the agent software stack, a 2,000 dollar annual license would generate 2 billion dollars in recurring annual software revenue, scaling to 10 billion dollars per year across 5 million GPUs. Huang emphasized that recurring software licensing will become an increasingly substantial pillar of Nvidia's business model alongside accelerator hardware sales.
The proposed agent software tier sits alongside Nvidia's existing AI Enterprise subscription, which lists at 4,500 dollars per GPU per year, underscoring Nvidia's push to capture recurring software margins from enterprise AI deployments.Industry analysts expect concrete adoption milestones to emerge over the coming quarters as initial deployments scale across production environments.
By bundling orchestration, telemetry, and guardrail microservices into a per GPU annual subscription, Nvidia is positioning its software platform as an enterprise operating system layer that scales directly with cluster size.
What to expect next
Cloud providers and enterprise buyers already facing steep GPU capital costs will weigh Nvidia's commercial agent software against open source inference and agent orchestration stacks such as vLLM, SGLang, and LangGraph.Industry analysts expect concrete adoption milestones to emerge over the coming quarters as initial deployments scale across production environments. If enterprise customers adopt the new agent software tier alongside Nvidia AI Enterprise licenses, software and services could grow into a much larger share of Nvidia's total data center operating income over the next three years.