Nvidia’s Huang Pegs Software at About $2,000 per GPU a Year

Nvidia chief executive Jensen Huang “put the price of licensing Nvidia’s software stack at about $2,000 per GPU annually,” The Register reported from a question-and-answer session at Nvidia’s GTC event in Berlin, in an article published on 22 October 2026. In plain words, that is a yearly fee for the software that runs on each Nvidia chip.

Nvidia’s own licensing guide lists its NVIDIA AI Enterprise software at $4,500 per GPU for a one-year subscription. By our arithmetic, Huang’s figure is about 44% of that list price, but the sources do not say whether the two refer to the same product.

Business Pill · OPERATING LEVERAGE

A short Business Engineer explainer on operating leverage: why software built once can serve 10 or 10,000 customers at about the same cost. It teaches the general idea only and says nothing about any company in this piece.

What Huang Said

• The price. According to The Register, Huang put the price of licensing Nvidia’s software stack at about $2,000 per GPU a year. The Register’s line is a paraphrase, not a direct quote.
• The setting. The Register reported the remark from a question-and-answer session at GTC Berlin, alongside Huang’s comments on AI safety, jobs and a $4,000 DGX Spark computer.
• On X. The account Wall St Engine posted on 23 October that Huang said “software licensing will become increasingly a major part of our company’s business” and that it could cost “about $2,000 per GPU per year” to run Nvidia’s software for AI agents in AI factories.

We did not find a transcript of the session on Nvidia’s newsroom or blog as of 08:30 UTC on 23 October; the words above come from The Register’s report and Wall St Engine’s post.

What Was Posted on X

Wall St Engine posted on 23 October that Jensen Huang said software licensing will become increasingly a major part of Nvidia’s business and could cost about $2,000 per GPU per year to run Nvidia’s software for AI agents in AI factories. The post’s revenue scenarios (500,000 licensed GPUs = $1 billion a year, up to 5 million = $10 billion) are the poster’s own napkin math, which it labels illustrative and not Nvidia guidance.

Nvidia’s Own List Prices

Nvidia’s licensing guide for NVIDIA AI Enterprise, its software suite for running AI in companies’ own data centres, publishes list prices for self-managed systems sold through its partner network. A subscription that includes support is $4,500 per GPU for one year and $18,000 for five years, which the guide marks as five years for the price of four. A perpetual licence is $22,500 per GPU with five years of support.

Nvidia’s list prices are for self-managed systems sold through its partner network. The sources do not say whether Huang’s figure refers to the same product or terms as the AI Enterprise list price. The per-year five-year figure is our arithmetic (18,000 / 5).

The guide also says the five-year subscription is included with NVIDIA H100, H200 NVL and A800 PCIe GPUs. As we read it, the gap between $4,500 and about $2,000 can have several explanations the sources do not settle: a different bundle, a per-year rate on a longer term, or a rounded figure in a live answer.

The Poster’s Math

Wall St Engine added what it called napkin math at that price, before discounts: 500,000 licensed GPUs would bring $1 billion a year, 1 million would bring $2 billion and 5 million would bring $10 billion. The poster labelled these “illustrative revenue scenarios using Jensen’s pricing example, not NVIDIA guidance or profit estimates.” The sources we read do not say how many GPUs carry a software licence.

Connecting the Coverage

Our Vera Rubin piece reported that on stage at the GTC Berlin keynote on 21 October 2026, Huang said Vera Rubin is another 30 times beyond Blackwell, while Nvidia’s release put the figure as 30x higher throughput per megawatt for agentic workloads. What’s new: a price for the software layer, not the chip.

Our Agent Compute Engine piece reported that Google Cloud said its open-source Agent Substrate runtime delivers as much as 10x higher compute density for AI agents. Our selloff piece reported that the Philadelphia SE Semiconductor Index plunged 6.7% on 22 October, according to Reuters. Our proofs piece reported Theorem’s cost model of about $40 per kilobyte of machine code for AI-written proofs.

The Structural Read

As we read it, a per-GPU yearly fee is a different kind of revenue from selling the chip: Nvidia’s guide prices AI Enterprise per GPU, per year, as a subscription that includes support.

As we read it, the two figures should not be read as a discount until the sources say they describe the same product: the guide lists $4,500 for one year and $18,000 for five years, and Huang’s figure was reported as about $2,000 for the software stack.

As we read it, Wall St Engine’s scenarios show the scale of the idea, not a forecast: the poster labelled them illustrative and not Nvidia guidance.

Three Implications

FOR BUYERS OF NVIDIA SYSTEMS Nvidia’s guide lists AI Enterprise at $4,500 per GPU for one year and includes a five-year subscription with H100, H200 NVL and A800 PCIe GPUs.

FOR NVIDIA’S BUSINESS MIX The Register reported Huang’s software figure alongside Nvidia’s $215.9 billion of revenue for fiscal 2026; the sources do not say how much of that is software.

FOR READERS OF THE $2,000 FIGURE The Register’s line is a paraphrase, and we did not find a transcript as of 08:30 UTC on 23 October.

The Business Engineer Lens

This story maps onto the Business Engineer framework Enterprise AI: From Software to Substrate.

The framework’s starting point: “Enterprise vendors don’t sell software anymore—they sell irreversibility.”

As we read it, a yearly software licence could be a way to keep earning from a chip after it is sold, on the GPUs that carry one; the sources do not say how many do. The framework was published in December 2025 and does not discuss Nvidia’s pricing.

What Is Not Established

The Register’s report paraphrases Huang; we did not find a transcript as of 08:30 UTC on 23 October, and Wall St Engine’s quotes are its own account. The sources do not say which products the $2,000 figure covers, whether it is a list price or a typical paid price, or how many GPUs are licensed. Nvidia’s list prices come from its own licensing guide and may differ from what partners charge. We did not contact Nvidia. Nothing here is investment advice.

The Bottom Line

Huang put Nvidia’s software licensing at about $2,000 per GPU a year at GTC Berlin, The Register reported. Nvidia’s own guide lists its AI Enterprise subscription at $4,500 per GPU for one year; the sources do not say whether the two figures describe the same thing.