How to activate NVIDIA AI Enterprise: licensing portal, NGC API keys and the H200 NVL subscription

How to activate NVIDIA AI Enterprise

To activate NVIDIA AI Enterprise (NVAIE), you register the entitlement NVIDIA issued for your order and then give the right people access to NVIDIA’s account services. For the subscription included with each H200 NVL, the entitlement is tied to the GPU serial numbers, which you enter on NGC under Organization, Activate Subscription. For a subscription bought per GPU, it is the NVIDIA Entitlement Certificate attached to the order confirmation, which you redeem by registering an NVIDIA Enterprise Account. After that you create NGC API keys to pull containers and NIM microservices, and you set up a licence server in the NVIDIA Licensing Portal only where VMs use vGPU for Compute.

NVIDIA’s Quick Start Guide for AI Enterprise infrastructure release 8, updated on 2 September 2026, puts these services behind one sign-in, the NVIDIA Application Hub, which leads to NGC, the Enterprise Support Portal and the Licensing Portal.

NVIDIA AI Enterprise Quick Start Guide and Licensing Guide (both updated 2 September 2026), NVIDIA activation instructions (document 2843294, October 2025), NGC user guide (updated 5 October 2026). The access column is our suggestion for a company of 500 to 2,000 staff.

Activating the subscription included with H200 NVL

NVIDIA’s licensing guide, updated on 2 September 2026, states: “Each NVIDIA H200 NVL Tensor Core GPU includes a five-year NVIDIA AI Enterprise subscription.” The term starts on the ship date of the GPU board to the OEM partner plus 90 days, and the guide adds that this start date “cannot be modified, as it is tied to the specific card”. The software may be installed “on the NVIDIA Certified Systems in which the GPUs are installed”, “but not on a different server or instance”. Activation does not move the start, so a card activated a year after delivery keeps the same end date, and the term may already be running when the server arrives.

The serial numbers come first. NVIDIA’s note on locating them (version 3, October 2023) names the rear plate near the PCI bracket, the output of nvidia-smi -a, and a short list printed with --query-gpu=serial and --format=csv,noheader. Some servers also show the GPU serials in the BMC. Our H200 NVL installation checklist records them before the cards go in. NVIDIA’s activation instructions (document 2843294, October 2025) then come down to four steps:

1. Sign in to NGC with a business e-mail address; if you have no NGC account, you are prompted to create one.
2. In the left-hand menu under Organization, choose Activate Subscription, fill in the company details and enter the serial numbers at the bottom of the form, separated by commas or new lines.
3. Click Activate Subscription, then Request Subscription; NVIDIA AI Enterprise appears on the subscription page as Active, or as Pending while approval runs, for which NVIDIA allows 24 to 48 hours.
4. Within 24 hours an e-mail from NVIDIA starts the Enterprise Account; it explains how to create the Enterprise Support account and how to reach the Licensing Portal.

Once approved, the software sits under Catalog, Enterprise Catalog in NGC. For two servers with four H200 NVL each, the form takes all eight serial numbers in one request. NVIDIA enables subscriptions on enterprise organisations, and the NGC user guide lists the Activate Subscription portal among the ways one is created. Unlike the individual organisation a user receives on signing up alone, an enterprise organisation “supports additional users, NGC teams, and role-based access rules.” Submit the request from the account that will own the organisation for the company, because “All org owners must share the same email domain.”

We supply the H200 NVL with its five-year NVIDIA AI Enterprise subscription stated in the quote, on one EU contract and invoice. Tell us how many cards and servers you plan and whether VMs will share them.

Activating a subscription bought per GPU

The L40S, L4 and RTX PRO cards come without AI Enterprise, so their subscriptions are bought per GPU, as explained in our NVIDIA AI Enterprise licensing guide. Activation starts from a document. In the Quick Start Guide’s words, “After your order is processed, you get an order confirmation with an attached NVIDIA Entitlement Certificate.” The NGC user guide names the e-mail “NVIDIA Entitlement Certificate - Ref.”, and NVIDIA asks anyone who is not an administrator to forward it to the data-centre administrator.

The Register link in the certificate’s instructions opens the NVIDIA Enterprise Account Registration form, and the account becomes active after the password is set and the e-mail address verified. A customer with an existing account logs in instead, and an evaluation account is linked by registering with the same e-mail address. During registration, the NGC user guide has you select or create an NVIDIA Cloud Account (NCA). A new NCA also creates an NGC organisation linked to it, while an NCA you already own can be selected so that the subscription lands in the company’s existing organisation. If the NCA belongs to a colleague, the guide says to contact its owner.

From the certificate e-mail to the first container pull, the order is:

1. Forward the certificate to the administrator who will own the NGC organisation.
2. Register or log in through the certificate’s link, enter the primary contact and choose the NCA, the company’s existing one or a new one.
3. Log in from the NVIDIA Application Hub e-mail and open the NVIDIA NGC card; the Subscriptions page shows the active NVIDIA AI Enterprise subscription.
4. Invite the team on the organisation’s Users page and assign their roles.
5. Generate a Personal key with the NGC Catalog service included, then run docker login nvcr.io with the username $oauthtoken and the key as the password.
6. Pull the first container from the enterprise catalogue and note the image tag in the change record.

The licensing guide states: “The subscription includes the software license and production level support services throughout the subscription.” It names this level Business Standard, and Business Critical is an upgrade NVIDIA sells on top. For the H200 NVL subscription, support cases go through the Enterprise Support account created from NVIDIA’s e-mail.

NGC organisation and API keys

NGC has two kinds of key. The private registry guide, updated on 5 October 2026, says “Any NGC org user can generate a personal key”, up to eight per user, and that “If the user is removed from the NGC org, the key’s validity is revoked.” Service keys work the other way: “The lifecycle of service keys is linked to the NGC org account, not associated with an individual user.” Only organisation owners and user admins manage them, an organisation can hold up to 64, and NVIDIA recommends them “when you require automated communication between machines”. NGC shows a key once, in NVIDIA’s words “NGC does not save your key, so store it securely.”

As of 5 October 2026 the private registry guide reads “NVIDIA plans to deprecate the legacy NGC API key after 2025.” NIM for LLMs does not accept it, and its prerequisites page, updated on 6 October 2026, states “Legacy API keys are not supported by NIM LLM and VLM.” Most NIM containers and models on NGC now download without a key. A Personal key with at least NGC Catalog among its services is needed for Production Branch models, for NIMs released before version 2.0.10 and for private or gated images, and it is passed to the container in the NGC_API_KEY variable. Access needs the NVIDIA Developer Program or an AI Enterprise licence. GPUs, drivers and memory for each NIM are covered in our guide to NIM requirements.

On a Kubernetes cluster built from two H200 NVL servers, the image pull secret should hold a service key, while each data scientist pulls with a personal key. When someone leaves, removing the user revokes their keys, and the cluster keeps pulling.

When a licence server is needed

NVIDIA’s licensing guide says the NVIDIA License System “is only required to be installed and configured when using NVIDIA vGPU for Compute drivers”, and that “NVIDIA AI Enterprise software runs with or without a valid license server connection.” A bare-metal H200 NVL server running containers therefore needs no further licensing step after NGC. Where VMs use C-series vGPU profiles, the Quick Start Guide adds: “To activate an NVIDIA vGPU for Compute, software licensing must be configured for the vGPU VM client when booted.”

That work happens in the NVIDIA Licensing Portal, which, according to the NVIDIA License System user guide, “provides access to the entitlements that you purchased and the licenses that they contain.” There you create a licence server, allocate licences to it and serve them from a Cloud License Service (CLS) instance hosted on the portal or a Delegated License Service (DLS) instance on your own network; each VM then receives a client configuration token. Setup, high availability and failure behaviour are in our article on the vGPU licence server with DLS or CLS. Further people are added as registered contacts of the organisation linked to the NVIDIA Enterprise Account, and the portal gives them role-based access.

Who should own the portal, NGC and support accounts

Assign the accounts to roles, since they outlive the people who set them up. The NGC roles include Org Owner, User Admin, Registry Admin, Registry User, Registry Read and User Read, and the NGC user guide shows an NVIDIA AI Enterprise Viewer role in an example of roles assigned by identity provider rules. Registry Read “can download and pull artifacts”, which is enough for most developers. A User Admin “can only grant roles that they possess”, so the person who onboards staff needs the roles they hand out.

Role names from NVIDIA’s NGC user guide and private registry guide (both updated 5 October 2026) and the NVIDIA License System user guide; the split of owners is our suggestion, not an NVIDIA requirement.

As an example, take a company with 1,200 staff that runs two inference servers with four H200 NVL each and one VM host with four L40S on C-series vGPU profiles. The eight H200 NVL serials go into one NGC request, the L40S certificate can be redeemed by selecting the NCA that holds that organisation so that both subscriptions sit in one place, and only the L40S host needs a licence server.

Our engineering partner Vixen.UNO designs, deploys and supports such a platform, under the same contract if you want it. Describe your hypervisor, the cards and how many VMs share each one in the form below.

Common activation problems

H200 NVL serials are left unrecorded until the cards sit in a racked server, or the request is submitted by an engineer who stays the only org owner and later leaves the company. A Pending status is taken for a failure, although NVIDIA allows 24 to 48 hours, or NVIDIA’s e-mail lands in the junk folder, which the instructions themselves warn about.

A Production Branch NIM fails at the first pull because the key is a legacy one, or a pipeline stops when a departed colleague’s personal key is revoked. A licence server is planned for bare-metal servers that do not need one. A further problem is planning the included subscription as five years from activation, although NVIDIA counts it from the board’s shipment plus 90 days.

What we supply

We supply NVIDIA AI Enterprise and vGPU licences together with the cards they belong to, the H200 NVL, L40S, L4 and RTX PRO 6000 Server Edition among them, on one EU contract and invoice. Where a card needs a licence for the job, it is in the quote, and for the H200 NVL the quote states the included five-year subscription. Details are on our NVIDIA AI Enterprise and vGPU licences page. We build AI servers to order around these cards with manufacturer warranty, with configuration and quote within one business day, and our engineering partner Vixen.UNO deploys them if you want it.

FAQ