NVIDIA AI Enterprise licence: pricing, deployment options, terms and who needs it

NVIDIA AI Enterprise licence: what it includes

NVIDIA AI Enterprise is licensed per physical GPU and sold as a subscription of one to five years, as a perpetual licence bought with five years of support or, in cloud marketplaces, per GPU per hour; pricing follows the GPU count, the term and the support level. Each H200 NVL includes a five-year subscription. The DGX Spark has a licence of its own, NVIDIA AI Enterprise for DGX Spark, which other AI Enterprise entitlements do not automatically include; our DGX Spark licence guide explains it. The licence covers a software platform organised in two layers. The application layer holds NIM microservices, NeMo, Triton Inference Server, TensorRT and TensorRT-LLM, RAPIDS, the PyTorch container, Riva, DeepStream, Holoscan, TAO, Omniverse and NVIDIA’s pre-trained models, released in feature, production and long-term branches. The infrastructure layer holds the data-centre driver, the vGPU manager and guest driver, Fabric Manager, the GPU, Network, DPU and NIM operators for Kubernetes, Base Command Manager, the container toolkit and Run:ai, which NVIDIA’s documentation of 10 August 2026 lists as included both self-hosted and as SaaS.

Three things are exclusive to licence holders, according to NVIDIA’s own licensing guide: select NIM containers, the production and long-term support branches, and vGPU for Compute. The fourth is enterprise support with response targets. Nearly everything else in the list also exists in open-source or freely downloadable form; self-hosted Run:ai is the exception, and NVIDIA’s free trial leaves it out. Our AI Enterprise page covers the licences we supply; this article explains how they are counted and when you can do without them.

How NVIDIA AI Enterprise is licensed: per GPU

A server or workstation that runs any AI Enterprise software needs a licence for every GPU installed in it; NVIDIA’s licensing guide says the licence is “required for every GPU installed on the server or workstation”. Boards with several GPUs count each GPU, so an A16 needs four licences. A CPU-only server or instance running the CPU parts of the stack needs one licence per server, regardless of how many processors it has. The licensing guide counts physical GPUs and does not mention MIG, so a card split into four MIG instances still needs one licence.

Response targets per NVIDIA’s Enterprise Support and Services guide; NVIDIA’s support matrix states that the listed GPUs and platforms “are supported with servers listed on the NVIDIA-Certified Systems”.

Two conditions apply. NVIDIA’s AI Enterprise 8.2 support matrix (2 September 2026) states that the GPUs and platforms it lists “are supported with servers listed on the NVIDIA-Certified Systems”, so check that the server model is on that list before you count on NVIDIA support. And a five-year subscription does not extend the life of the software branch you deploy; the branch keeps its own end-of-support date.

Which GPUs and systems include AI Enterprise

Some GPUs include a subscription in the price and others include nothing, so check this table before comparing the L40S and the H200 NVL on card price alone.

Per NVIDIA’s licensing guide and product pages, as of October 2026; SE: Server Edition. If a quote bundles AI Enterprise with the RTX PRO 6000 Server Edition, it is a purchased NVIDIA subscription, not an included one: server makers list a per-GPU NVIDIA software kit for this card, and NVIDIA’s product terms restrict such platform subscriptions to the platform they are sold for. Have its term stated.

The H200 NVL row changes comparisons the most. If a project needs AI Enterprise anyway, for vGPU compute profiles or for supported NIM in production, each H200 NVL carries five years of it. The RTX PRO 6000 comparison then has to add eight licences to an eight-card server before the totals can be compared.

NVIDIA AI Enterprise pricing and deployment options

NVIDIA sells AI Enterprise per GPU through two channels, which its licensing guide, last updated on 2 September 2026, describes. For self-managed systems, on-premise or at the edge on NVIDIA-Certified servers, NVIDIA’s pricing page lists the licence as available through the NVIDIA Partner Network, as a subscription of one to five years or as a perpetual licence with five years of support. For cloud-hosted systems, it is offered in the marketplaces of the large public clouds, where NVIDIA says it “is priced per GPU per hour on an on-demand/pay-as-you-go basis”.

Per NVIDIA’s AI Enterprise licensing guide and pricing page, both last updated on 2 September 2026; the cloud provider bills the instance itself.

Some hardware comes with the licence. Each H200 NVL includes a five-year subscription, as do the H100 PCIe and H100 NVL, and NVIDIA starts the term 90 days after the card ships to the server maker, a date that cannot be changed. DGX systems with Hopper GPUs carry AI Enterprise in the DGX software bundle; for DGX systems with Blackwell GPUs, the guide states that the licences “need to be purchased separately”. The DGX Spark has a product of its own, which our guide to NVIDIA AI Enterprise for DGX Spark covers.

The price of an AI Enterprise licence depends on the number of physical GPUs, on subscription or perpetual, on the term and on the support level, and a GPU that already includes a subscription needs no second one for that term. NVIDIA’s pricing page publishes per-GPU list prices in US dollars for each term and prices the five-year subscription at the rate of four years. A CPU-only server counts one licence per server, however many processors it has. In the cloud, the hourly charge comes in addition to the instance, and the guide notes that not all components are available on every cloud deployment.

We quote AI Enterprise per GPU with the hardware it runs on, on one EU contract and invoice. Tell us your GPUs and the term you need, and we reply within one business day with a written quote.

vGPU for Compute: the only part the driver enforces

Most of the platform runs whether or not a licence server can be reached: NVIDIA’s own words are that the software “runs with or without a valid license server connection”, and the licence system is only required when vGPU for Compute drivers are in use. That case is enforced in the driver, which is why it matters most to virtualisation teams.

C-series profiles, the compute vGPU types such as L40S-48C or DC-8C on the RTX PRO 6000 Server Edition, are licensed only through AI Enterprise. The RTX vWS, vPC and vApps licences that virtual-desktop teams already own do not cover them. Each C-series vGPU assigned to a VM checks out a licence at boot and returns it at shutdown, and NVIDIA counts the licence per physical GPU. Have the quote confirm the count before you design a card with many small compute vGPUs.

The licence server itself comes in two forms. The Cloud License Service is hosted on NVIDIA’s licensing portal and needs outbound HTTPS. The Delegated License Service is a virtual appliance on your network, clusterable up to nine instances, with node-locked licences for air-gapped hosts available since vGPU software 15.0. Clients carry a client configuration token generated on the service instance. In daily operation, the more likely problem is a firewall change that cuts the VMs off from the licence server, which, on the cloud service’s default, becomes visible only seven days later.

What runs without an AI Enterprise licence

The CUDA toolkit is free under its EULA. TensorRT-LLM and vLLM are licensed under Apache-2.0, Triton Inference Server under BSD-3, and NeMo is open source. A single-tenant inference server running vLLM on eight RTX PRO 6000 Server Edition cards needs no AI Enterprise licence to run, and many such servers run without one. Which NVIDIA tools are free for commercial use, tool by tool, is in our article CUDA, TensorRT, NIM: free or NVIDIA AI Enterprise.

Without the licence, NIM containers are available to developers for research and testing on up to 16 GPUs, and NVIDIA’s NIM FAQ, last modified on 6 August 2026, states: “Using NIM in production requires an NVIDIA AI Enterprise license.” NVIDIA’s NIM offerings page of 6 October 2026 lists the NIM Certified Feature Branch “at no charge under the applicable license terms” and the Production Branch as requiring a subscription, so have the quote state which offering a production server rests on. You also get the monthly feature branch instead of a production branch with nine months of stable APIs, and no NVIDIA support contract for a driver bug in production. A 90-day trial licence exists for the move to production; it needs an NVIDIA-Certified server and excludes Run:ai.

With one or two GPUs, one team, an open-source serving stack and no VMs sharing the card, the licence is optional and mainly buys support. With C-series compute vGPUs shared by several teams in virtual machines, or NIM in production, it is required, and an eight-card server needs eight licences, regardless of the card price.

We supply AI Enterprise and vGPU licences and quote them with the hardware. If you are not sure which case is yours, describe your setup and we will tell you whether the licence is needed.

Per user or per GPU: the two licence families

The same card can carry two different licence models depending on the profile type, and mixing them up is the most common mistake we see on quotes.

NVIDIA’s vGPU licensing guide lists CUDA and AI development under RTX vWS for Q-series desktops; C-series compute vGPUs are licensed only through AI Enterprise, so a Q-series and a C-series profile on the same card are two different licences.

Our guide on MIG and vGPU counts covers how many VMs each card holds; this table shows which licence, and so which budget line, each VM needs.

Release branches and end-of-support dates

AI Enterprise separates the infrastructure layer from the applications, and each has its own branch cadence. On the application side, feature branches ship monthly with one month of support and may change APIs, production branches ship every six months with nine months of support and monthly security patches, and long-term support branches arrive every 30 months with three years of support and quarterly patches. Infrastructure releases are supported for one year, or three when NVIDIA designates a release as long-term, which it does at least once every two years and at least once per GPU architecture.

Per NVIDIA’s AI Enterprise lifecycle pages and the planned end-of-life dates in its AI Enterprise documentation, as of October 2026. NVIDIA’s driver table ends the R595 and R580 driver branches a month earlier, in March 2027 and June 2028. Infrastructure 8.0 dropped V100, RTX A4000, RTX 4000 SFF Ada and the Quadro RTX 8000, 6000 and 4000 from vGPU for Compute; they stay supported on 7.x.

A regulated environment that needs three-year API stability belongs on the long-term branch on infrastructure 7.x. The long-term application branch reaches end of support in October 2027 and infrastructure 7.x in July 2028, as planned by NVIDIA, so the upgrade should already be in the plan. The Triton container with the TensorRT-LLM backend is no longer in production branch 6; deployments move to the vLLM backend variant.

Where VMware, Red Hat and Nutanix put the licence

Broadcom’s requirements for VMware Private AI Foundation with NVIDIA on VCF 9.1, updated on 21 September 2026, list three licences: the VCF subscription, a Private AI Foundation add-on licence and, for vGPU, NVIDIA AI Enterprise for the ESX host driver, the guest drivers and the NGC container images, plus a licence server or portal token. Broadcom’s VCF 9.1 licensing documentation adds that since 3 November 2025 a VCF subscription may also include a Private AI Foundation licence, so have the VMware quote confirm whether a separate line applies. For DirectPath GPUs, Broadcom documents access without the AI Enterprise requirement, though AI Enterprise software running on them still needs its per-GPU licence. Broadcom’s minimum is three GPU-enabled ESX hosts in the initial cluster of the workload domain. Red Hat OpenShift AI enables NIM with an NGC key from an account that carries the AI Enterprise Viewer role, so the entitlement comes from NVIDIA, not from Red Hat. Nutanix Enterprise AI consumes NIM with your own NIM token and states that NIM is licensed through AI Enterprise. In every case the per-GPU licence is a separate line from the platform, and the VMware core count is a separate line again.

Common licensing mistakes we come across

Cards are ordered without the licence, and VMs on vGPU compute profiles drop to idle-level compute after 20 minutes; the fix is a per-GPU purchase, not a per-VM one. VMs are counted instead of GPUs, or the vWS licences already in the environment are assumed to cover C-series compute vGPUs. A board with several GPUs is counted as one. Included subscriptions are never activated because nobody registered the serial numbers. A DGX Spark is bought on the assumption that the department’s AI Enterprise licence covers it, although NVIDIA says other entitlements do not. The hypervisor is outside the support matrix: version 8.2 lists ESXi, Red Hat and Ubuntu KVM and Nutanix AHV, and Proxmox, Hyper-V and XenServer are not in the table. And AI Enterprise includes the DOCA driver for networking on BlueField adapters, so have the quote say whether the adapters need support of their own.

What we supply

We supply NVIDIA AI Enterprise subscriptions and perpetual licences across the EU, on one EU contract and invoice with the GPUs they belong to. If you tell us the use case and the GPUs, we recommend the edition and term and flag what you do not need. For the H200 NVL the quote states the five-year subscription that comes with each card; for RTX PRO 6000 servers we derive the licence lines from the vGPU plan: AI Enterprise per GPU for compute profiles, vWS per user for desktops. Prices depend on the form, the GPU count and the term, so they come with the quote.

FAQ