Nvidia AI Software Licensing Fee: The $2,000 per GPU Figure
Jensen Huang says running Nvidia's AI agent software costs about $2,000 per GPU a year. It is not an official price yet. Here is what he said, what Nvidia already charges and what it means for buyers.
- At GTC Berlin on October 22, 2026, Jensen Huang said running Nvidia's AI agent software costs about $2,000 per GPU per year.
- It is not an official price: Nvidia has not said what it covers, when it starts or how it relates to AI Enterprise at $4,500 per GPU a year.
- Data center operators would pay, and how Nvidia counts a GPU (die or module) could matter as much as the price.
Nvidia has not formally launched a new fee. What happened is that CEO Jensen Huang, answering questions at Nvidia's GTC Berlin conference on October 22, 2026, said it costs about $2,000 per GPU per year to run all of Nvidia's software for AI agents, and that software licensing will become a bigger part of the business. As of October 26, Nvidia has not said which products that covers, when it would apply or how it relates to the $4,500-per-GPU yearly price it already lists for its AI Enterprise suite.
The remark spread quickly because it hints at how Nvidia wants to earn recurring revenue from the millions of GPUs it ships. Here is what was said, who pays for Nvidia software today and what it could mean for buyers.
What Jensen Huang actually said
The Register first reported the figure in a story on Huang's question-and-answer session at GTC Berlin. It paraphrased him as putting the cost of licensing Nvidia's software stack at roughly $2,000 per GPU a year.
Fuller wording appeared in later coverage. Notebookcheck quotes Huang saying, "Software licensing will become increasingly a major part of our company's business," and that "it's about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents." Startup Fortune carries the same first sentence and describes the number as a ballpark figure rather than an official price.
Three details matter. Huang offered the number as an example of how Nvidia licenses software, not as a price list. He tied it to software that runs AI agents in what Nvidia calls "AI factories," meaning large AI data centers. And the figure was not part of his main keynote the day before. The keynote transcript covers agents, new hardware and data center economics, but no per-GPU software pricing.
What Nvidia already charges for software
Nvidia already sells software per GPU. Its official licensing guide, last updated September 2, 2026, lists these prices for NVIDIA AI Enterprise, its suite of AI frameworks, inference tools and support, on self-managed systems:
- $4,500 per GPU for a one-year subscription with support
- $18,000 per GPU for five years, which Nvidia describes as five years for the price of four
- $22,500 per GPU for a perpetual license with five years of support
- $1,125 per GPU for one year for qualifying education, research and startup-program customers, a quarter of list price
- $1 per GPU per hour, plus cloud instance costs, through the AWS, Azure and Google Cloud marketplaces
The guide also notes that a five-year subscription comes bundled with certain GPUs, including H100, H200 NVL and A800 PCIe cards.
That is why the $2,000 figure confused people: it is less than half the one-year AI Enterprise price. Notebookcheck, VGTimes and Startup Fortune all caution against reading it as a price cut, because Nvidia has not said whether the two numbers cover the same software or terms.
Who would pay, and how much
Any fee like this would land on organizations that run Nvidia data center GPUs for AI: cloud providers, GPU-rental specialists known as neoclouds, large enterprises and AI labs. Huang's comments were about data center software for agents, not consumer graphics cards.
The scale is what makes the number notable. VGTimes and Startup Fortune point out that every million licensed GPUs at $2,000 would add up to about $2 billion a year. Notebookcheck's own example goes further: five million paid licenses would mean $10 billion a year. These are outside back-of-envelope calculations, not Nvidia forecasts.
For context, Nvidia reported fiscal 2026 revenue of $215.9 billion, of which $193.7 billion came from its data center business, and $96.2 billion in its most recent reported quarter. Those releases do not break out software revenue.
For a single server, the math is simpler. The Register's 2025 analysis used an eight-GPU HGX B200 system, which costs $36,000 a year in AI Enterprise licenses at $4,500 per GPU. At $2,000 per GPU, the same eight GPUs would cost $16,000 a year, if the fee applied on the same basis.
The fine print: what counts as a GPU
How Nvidia counts GPUs could matter as much as the price. In April 2025, The Register reported that Nvidia had started counting individual chip dies, rather than whole modules, as GPUs on its air-cooled HGX B300 systems. Under that convention, an eight-module HGX B300 is rated as 16 GPUs, so per-GPU licensing would double if prices stayed the same. At the time, Nvidia said pricing details for B300 were still being finalized.
Startup Fortune raises the same question about the new figure: it is not clear whether a "GPU" means a die or a module. Nvidia has not answered publicly.
Why Nvidia is pushing software now
The comment came during a week focused on AI agents. On October 21, Nvidia announced the OpenShell Reference Architecture, a design for running agents inside set limits. It pairs Vera Rubin systems for model inference, Vera CPUs for agent tools and BlueField-4 data processing units running a security layer called Nvidia Sentry. Partners named in the announcement include model makers Anthropic, OpenAI, Google and Meta, and cloud providers such as CoreWeave, Nebius and IREN.
Nvidia's developer blog says Sentry, which monitors agent behavior and can quarantine a misbehaving agent within milliseconds, has planned availability early next year. The OpenShell runtime itself is open source, Huang said in his keynote.
A per-GPU license would turn that agent stack into recurring revenue. Startup Fortune argues that subscriptions would smooth out Nvidia's cyclical hardware sales and add another layer of customer lock-in on top of its CUDA developer ecosystem.
Reactions so far
Most reaction has focused on the missing details. LAVX News warned that buyers should not treat the $2,000 figure as a confirmed price cut. Notebookcheck and VGTimes listed the open questions: which products are covered, when it would start, how many GPUs would qualify and how it overlaps with AI Enterprise. As of October 26, 2026, none of the coverage reviewed for this article reported a clarification from Nvidia or a named analyst note on the comment.
What it means for buyers
For now, nothing changes on invoices. Nvidia's published list prices remain the reference point, and the $2,000 figure is not on any price list. Still, organizations planning AI clusters can take a few general steps:
- Budget for software as a recurring per-GPU cost, not a one-time purchase.
- Ask Nvidia partners how licenses are counted for the exact system, especially multi-die Blackwell and Rubin hardware.
- Check whether your GPUs include bundled AI Enterprise terms.
- Compare the $1-per-GPU-hour cloud option with an annual subscription based on real usage.
This is general information, not purchasing advice. Nvidia's licensing guide and its authorized partners are the official sources for current pricing.
Our take
Based on the reporting, many headlines overstate what happened. Huang gave an illustrative figure in a Q&A, not a product launch. The direction, though, is clear: Nvidia wants agent software, security and orchestration to become a paid layer on top of its hardware, and OpenShell plus Sentry give it something specific to sell.
The bigger risk for buyers may be ambiguity rather than the $2,000 itself, particularly around how GPUs are counted and what is bundled. Neoclouds, covered in our neocloud explainer, would feel any change most directly. Always-on agent apps such as Anthropic's new Claude Duet show the kind of workload Nvidia is targeting.
What to watch next
- An official Nvidia price list or product page for agent software licensing.
- Comments on software revenue in Nvidia's next quarterly results.
- Sentry's planned arrival in early 2027.
- Any update to Nvidia's licensing guide on how dies and modules are counted.
Frequently asked questions
Is Nvidia charging $2,000 per GPU for AI software?
Not officially, as of October 26, 2026. Jensen Huang gave the figure as an example during a Q&A at GTC Berlin on October 22. Nvidia has not published a price list, start date or product scope for it.
How much does Nvidia AI Enterprise cost?
Nvidia's licensing guide lists AI Enterprise at $4,500 per GPU for a one-year subscription, $18,000 per GPU for five years and $22,500 per GPU for a perpetual license with five years of support. Cloud marketplace use is $1 per GPU per hour plus instance costs. Qualifying education and startup-program customers pay a quarter of list.
Who would pay Nvidia's software licensing fee?
Organizations running Nvidia data center GPUs for AI, such as cloud providers, neoclouds, enterprises and AI labs. Huang's comments concerned software for running AI agents in data centers, not consumer graphics cards.
Why does it matter how Nvidia counts a GPU?
Licenses are priced per GPU. The Register reported in 2025 that Nvidia began counting each die, rather than each module, as a GPU on HGX B300 systems, which would double the license count for the same hardware if prices stayed the same. Nvidia has not said which method would apply to the new figure.
When would a new Nvidia software fee start?
Nvidia has not given a date. Related agent-security software, Nvidia Sentry, has planned availability in early 2027, according to Nvidia's developer blog. Check Nvidia's licensing guide or an authorized partner for current pricing.
Sources
- Nvidia's Huang: Don't mention the AI-pocalypse, The Register
- Nvidia discusses $2,000-per-GPU annual fee for running AI agents, Notebookcheck
- Nvidia Floats Charging $2,000 a Year Per GPU for Its AI Agent Software, Startup Fortune
- NVIDIA at GTC Berlin 2026 Keynote: Europe's AI buildout (transcript), Investing.com
- NVIDIA Enterprise Licensing Guide: Pricing, NVIDIA
- NVIDIA Eyes $2 Billion a Year From AI Software Licensing, VGTimes
- NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2026 (Form 8-K exhibit), NVIDIA via SEC EDGAR
- NVIDIA Announces Financial Results for Second Quarter Fiscal 2027 (Form 8-K exhibit), NVIDIA via SEC EDGAR
- Nvidia's AI suite may get a whole lot pricier, thanks to Jensen's GPU math mistake, The Register
- Nvidia GTC Berlin 2026: Nvidia Launches OpenShell Reference Architecture for Safe AI Agents, StorageNewsletter
- Inside NVIDIA Sentry: Secure AI Agents at Runtime, NVIDIA Technical Blog
- Nvidia floats $2,000 annual GPU software fee for agent stacks, pricing details murky, LAVX News
How we wrote this
GessyMorgan explainers are built from public reporting and official sources. This post draws on the 12 sources listed above, each linked where we use it. It was drafted with AI assistance and checked against those sources before publishing. We did not test products, attend events or interview anyone for this article. Facts are current as of October 26, 2026, and we update the post when they change.
