Nvidia floats $2,000-per-GPU yearly fee for agent software
Jensen Huang put Nvidia's software license at about $2,000 per GPU a year at GTC Berlin. Here's what it means for AI agent hosting costs and your vendor quotes.
Nvidia wants to be paid twice: once for the chip, then every year for the software that runs on it. At GTC Berlin on October 22, Jensen Huang put the price of licensing Nvidia's software stack at about $2,000 per GPU annually, The Register reported. An Nvidia per-GPU software fee sounds like a data center problem. It lands on your AI agent bill sooner or later.
What actually happened
The Register states the figure in its own words, not as a direct quote, and gives no detail on what the license covers. Other coverage ties it to software that runs agents in "AI factories." Nvidia has not published a price list, a product name, or a start date for this fee. Treat $2,000 as an order of magnitude from the CEO, not a quote you can plan against.
What we can check is the price Nvidia already publishes. Its AI Enterprise licensing guide lists:
- $4,500 per GPU for a one-year subscription on self-managed systems
- $18,000 per GPU for five years (priced as four)
- $1 per GPU per hour in the AWS, Google Cloud, Azure and Oracle marketplaces, plus the instance cost
So the floated figure is less than half the existing list price. Whether it is a new agent-specific tier, a bundle, or a different way to count the same thing, Nvidia has not said.
Why it matters for your business
You do not buy GPUs. Your vendors do. Most small businesses run agents through an API or a hosted platform. If Nvidia adds a recurring per-GPU software charge, cloud and inference providers pass it through in token prices or seat fees.
Software is the sticky part. A chip is a one-time purchase. A yearly license ties the stack to Nvidia's tools for as long as it runs. That is the lock-in to watch, not the hardware.
Ask what your quote assumes. When a vendor prices a private model deployment, ask whether Nvidia software licensing is in the number, and whether it is per GPU, per hour, or per year. A $1-per-GPU-hour line looks small until it runs 8,760 hours.
Keep the model layer swappable. If your agent code calls one provider's SDK everywhere, any upstream price change becomes your problem. A thin adapter between your app and the model lets you move.
Key takeaways
- Huang put Nvidia's software license at about $2,000 per GPU per year at GTC Berlin on October 22
- Nvidia has not named the product, the start date, or how it relates to existing licenses
- Nvidia AI Enterprise already lists at $4,500 per GPU per year, or $1 per GPU-hour in cloud marketplaces
- Recurring chip-vendor software fees reach you through your inference provider's pricing
- Build agents with a swappable model layer so upstream price moves stay optional
Worried your AI bill depends on someone else's price list? We build vendor-agnostic agent systems you own, with the model behind an adapter you can swap. See what we build, or check the math first with our ROI calculator.
Sources: The Register, NVIDIA AI Enterprise Licensing Guide.
- #nvidia
- #ai-agents
- #software-licensing
- #gpu-pricing
- #ai-costs
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
NVIDIA Nemotron 3.5 Bolt E4B: a 4GB local model to post-train
NVIDIA's Nemotron 3.5 Bolt E4B runs in about 4GB of VRAM with tool calling and 64K context. NVIDIA says post-train it before production. Here's why.
Read itNanbeige4.3-33B: Apache 2.0 open agent model, 512K context
Nanbeige4.3-33B is an Apache 2.0 open agent model with 512K context and big Terminal-Bench claims. The numbers are self-reported. Run your own evals first.
Read it