Skip to content

Back to the catalog

Nvidia discusses $2k-per-GPU annual fee for software
tomshardware.comarticle

Nvidia discusses $2k-per-GPU annual fee for software

Tom's Hardware reports that Jensen Huang put Nvidia's software licensing at about $2,000 per GPU per year for software that runs agents on AI factories, said at GTC Berlin on 22 October 2026. Nvidia AI Enterprise is listed at $4,500 per GPU per year, and Nvidia has not said which products or terms the new fee covers. The piece also covers the OpenShell Reference Architecture and Sentry agent monitoring on BlueField-4.

TL;DR Reports Huang's $2,000 per GPU per year agent software license figure, set against the $4,500 AI Enterprise price, with unclear scope and timing.
Document

What it claims

Jensen Huang said at GTC Berlin on 22 October 2026 that Nvidia's software licensing will be "about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories." He framed it as an example ("for example"), not a published price. Nvidia has not said when the fee starts, whether $2,000 is the standard rate, which products it covers, or how many GPUs it would apply to. Nvidia AI Enterprise is listed at $4,500 per GPU for a one-year subscription. Tom's Hardware says the lower figure should not be treated as a confirmed price cut, because Nvidia has not confirmed that the two prices cover the same software or terms.

The revenue figures come from FourWeekMBA, not Nvidia. Its scenario assumes five million paid licenses: $10 billion a year, or $50 billion over five years if every customer renews.

How it works

The fee sits alongside Nvidia's agent-security stack. Nvidia announced the OpenShell Reference Architecture on 21 October 2026. It combines Vera Rubin computing infrastructure, Vera CPUs and BlueField-4 DPUs running Nvidia Sentry.

  • OpenShell is a secure runtime for AI agents. It sets boundaries on what actions, tools and resources an agent can use.
  • Sentry runs on the DPU and watches agent activity independently of the agent. It can quarantine an agent that tries to cross those boundaries.
  • Where it runs: the controls live in the infrastructure, not only in safeguards inside the model or application.
  • Lineage: it builds on the Open Agent Safety Platform, which Nvidia introduced in September 2026, and extends that work to enterprise infrastructure and edge systems.
flowchart LR
  A[Agent action] --> B[OpenShell runtime]
  B --> S{Sentry on BlueField-4}
  S -->|within boundary| T[Tools and data]
  S -->|crosses boundary| Q[Quarantine agent]
  R[Access request] --> AP{Admin review in Slack}
  AP -->|approve| B
  AP -->|reject| X[Denied]

Integrations

Anthropic is integrating OpenShell and BlueField hardware with Claude Managed Agents. The aim is to give businesses more control over what agents can access inside execution sandboxes. Salesforce built a Slack Admin Bridge for OpenShell. In Slack, administrators can review agent access requests, approve or reject permissions, and monitor agent activity.

Limits

The price is a spoken remark from a video, not a price list. Before budgeting, check four things with Nvidia: whether it overlaps with an existing AI Enterprise subscription, which products it covers, when it takes effect, and whether OpenShell or Sentry fall under it. The article gives no OpenShell or Sentry availability dates, API details, performance overhead or pricing. It also does not say whether OpenShell needs BlueField-4 hardware.

Organisations named here

  • NVIDIA
  • Anthropic
  • Salesforce
  • Slack
  • Tom's Hardware
  • FourWeekMBA
  • BG2Pod
  • CoreWeave
  • GTC Berlin
  • Jensen Huang
  • Sentry
  • Vera Rubin

Technologies named here

  • Nvidia AI Enterprise
  • Nvidia OpenShell
  • Nvidia Sentry
  • Nvidia OpenShell Reference Architecture
  • Nvidia Open Agent Safety Platform
  • Claude Managed Agents
  • Slack Admin Bridge for OpenShell
  • OpenShell
  • BlueField-4
  • Slack
  • Claude
  • Open Agent Safety Platform
  • Vera Rubin

Techniques named here

  • Infrastructure-level agent security