NVIDIA AI Enterprise for DGX Spark: what the licence includes and how to activate it

What NVIDIA AI Enterprise for DGX Spark is

NVIDIA AI Enterprise for DGX Spark is a separate edition of NVIDIA AI Enterprise, licensed for the DGX Spark and for GB10 partner systems, and it is not included when you buy the hardware. NVIDIA introduced it in March 2026. It adds a validated software stack from NVIDIA’s NGC catalogue, enterprise support for that stack and, according to NVIDIA’s announcement, monthly updates for security and bug fixes. NVIDIA’s quick start guide for the edition (updated 10 September 2026) states that the entitlement “is available only if you purchased it”, requested an evaluation or received an NVIDIA Entitlement Certificate that lists it.

Older answers on NVIDIA’s developer forum predate the edition: a reply of 25 January 2026 reads “At this time, NVIDIA AI Enterprise is not offered for DGX Spark.” On 16 March 2026 NVIDIA announced the new edition on the same forum, as a version of NVIDIA AI Enterprise “designed specifically for developing and running production workloads on DGX Spark and GB10-based systems”. NVIDIA’s product-specific terms for its AI products, last modified on 15 April 2026, define DGX systems as NVIDIA-branded, “except for DGX Spark which may be NVIDIA or OEM branded systems”. The licence therefore applies to the Founders Edition and to the GB10 systems of NVIDIA’s partners alike.

How NVIDIA AI Enterprise is counted on servers, and which cards include it, is explained in our NVIDIA AI Enterprise licensing guide.

What the DGX Spark licence includes

NGC lists the software in a collection named DGX Spark, last updated on 16 March 2026, which describes NVIDIA AI Enterprise - DGX Spark as “an enterprise-grade software solution for deploying critical, production applications” on the DGX Spark and on GB10 partner systems. Under the heading “NVIDIA AI Enterprise - DGX Spark includes”, the collection names five NIM microservices: Multi-LLM NIM, NVIDIA-Nemotron-3-Nano-30B-A3B, NVIDIA-Nemotron-Nano-9B-v2, Llama-3.1-8B and Qwen3-32B. It adds PyTorch, RAPIDS, CUDA Deep Learning and the Holoscan SDK as frameworks, PyTorch, TensorRT and TensorRT-LLM as runtimes, the NIM Operator and GPU Operator for Kubernetes, and the GPU and ConnectX-7 drivers.

The scope is narrower than that of the server licence. NVIDIA’s solution overview for the Spark edition (March 2026) states: “The full license encompasses NIMs with larger models and other software targeting data center workloads.” In the same overview, NVIDIA enterprise support “is available during business hours in your local time zone”, with an option for support outside those hours.

For updates, NVIDIA’s announcement of 16 March 2026 lists “Monthly updates for security and bug fixes”. NVIDIA’s support page for DGX Spark (updated 10 September 2026) lists hardware support through tickets, chat and the DGX Spark forum without any entitlement condition, while enterprise technical support for the Spark software stack requires the NVIDIA AI Enterprise for DGX Spark entitlement.

What a DGX Spark runs without the licence

Without the licence, a DGX Spark is a complete development system. It runs DGX OS, NVIDIA’s Ubuntu 24.04-based operating system, with the GPU driver, CUDA 13.0, Docker and the NVIDIA Container Toolkit preinstalled. NVIDIA’s playbooks at build.nvidia.com/spark take you through Open WebUI with Ollama, vLLM, ComfyUI and Unsloth fine-tuning, and open-source engines such as vLLM run under their own licences, as on any server. Our DGX Spark getting started guide covers the first boot, and our DGX Spark specs the hardware.

NIM microservices also run without the licence for development. The NGC chapter of NVIDIA’s DGX Spark user guide points developers to membership of the NVIDIA Developer Program and an NGC Personal API key “for research, development, and experimentation”, and NVIDIA’s NIM FAQ sets the Developer Program limit at 16 GPUs. NVIDIA’s NIM playbook for the Spark (updated 30 September 2026) lists an NGC account with an API key as its prerequisite and does not mention NVIDIA AI Enterprise.

NVIDIA AI Enterprise for DGX Spark quick start guide, support page and NGC chapter of the DGX Spark user guide (all updated 10 September 2026), NGC DGX Spark collection (16 March 2026), solution overview 5004550 (March 2026), NVIDIA’s forum announcement (16 March 2026) and NIM FAQ, read on 16 October 2026.

When a company needs the licence

NVIDIA’s NGC chapter for the DGX Spark names the cases in one sentence: “For production use, enterprise support, or supported software lifecycle requirements, obtain or activate an NVIDIA AI Enterprise license.”

NVIDIA’s NIM FAQ states: “Using NIM in production requires an NVIDIA AI Enterprise license.” The same FAQ counts any use other than development, testing, research or evaluation as production. A Spark that serves an internal assistant to a department through a NIM container therefore needs the licence. The same model served by vLLM or Ollama needs no NVIDIA AI Enterprise licence, and the team then supports that stack itself or through a partner.

NVIDIA’s product-specific terms contain one exception for NIMs. Certain NIMs designated for RTX GPUs or “NVIDIA Grace-Blackwell client systems” may be used without a subscription if they run on a PC or workstation with NVIDIA RTX GPUs and are “not used in a commercial kiosk, server or other system used to service multiple users”. Whether a Spark shared by a team falls under that exception is a legal assessment for the company’s legal department.

A team that wants NVIDIA to take a case about a container or driver fault on the Spark needs the entitlement, and so does a change process that only admits software with a supported lifecycle.

We supply NVIDIA AI Enterprise for DGX Spark with new Spark systems or for the Sparks you already run, with the term stated in the quote. Tell us how many Spark systems you run or plan and which of them will serve production users.

How it differs from the per-GPU licence for servers

NVIDIA’s licensing guide (updated 2 September 2026) states: “A software license is required for every GPU installed on the server or workstation that will host any software that is included with NVIDIA AI Enterprise.” The Spark has one GPU, and reseller listings describe each Spark subscription as “1 GPU”, so one subscription covers one Spark.

Two rules keep the Spark edition and the server licence apart. Under section C.2 of NVIDIA’s product-specific terms, NVIDIA may offer subscriptions “solely for use on designated platforms, such as IGX, DGX Spark, or RTX PRO Server”, and a subscription bought for one of them is “restricted to use solely on that particular platform”. NVIDIA’s DGX Spark support page adds: “Other NVIDIA AI Enterprise entitlements do not automatically include this product or its support.” A Spark subscription does not move to a server, and a server licence does not automatically bring the Spark edition or its support.

NVIDIA product-specific terms for AI products (last modified 15 April 2026), NVIDIA AI Enterprise licensing guide (2 September 2026), DGX Spark support page (10 September 2026), NGC DGX Spark collection; reseller catalogue read on 16 October 2026.

As an example, take a company with 1,000 staff whose AI team develops on four DGX Spark and serves its models from two servers with eight RTX PRO 6000 Server Edition each. If the servers run NIM in production, they need 16 per-GPU server licences. The Sparks need no licence for development; if the team wants NVIDIA enterprise support on them, four DGX Spark subscriptions come on top as a separate product. With H200 NVL cards in the servers instead, each card brings its five-year subscription, and any Spark subscriptions remain lines of their own.

We size the licences with the hardware, for the Spark systems and for the GPU servers, on one EU contract and invoice. Send us the number of Sparks and server GPUs through the form below, with the term you have in mind.

Subscription terms and the 90-day evaluation

The NVIDIA pages for the Spark edition give no term table. NVIDIA’s product-specific terms allow subscriptions for designated platforms such as DGX Spark to come with terms that differ from the full subscription, so the term is a line to confirm in the quote. A US reseller catalogue read on 16 October 2026 lists a three-year subscription with Business Standard support for one GPU, and renewals of one to five years, under NVIDIA part numbers that begin with 731-AI7030. We found no perpetual licence for the Spark edition, while NVIDIA’s licensing guide describes a perpetual licence for servers that “includes five-year Business Standard Support services”.

For a test before purchasing, the quick start guide states that “NVIDIA offers a 90-day trial of NVIDIA AI Enterprise”. Registration runs on NVIDIA’s enterprise product registration site. The solution overview adds: “The 90-day license only provides community-driven support.” Enterprise support therefore starts with a bought subscription. Our guide to buying a DGX Spark in the EU puts the question of enterprise support on its checklist before the order.

How to activate NVIDIA AI Enterprise for DGX Spark

The quick start guide says “After your order is processed, you will receive an order confirmation message from NVIDIA”, with the NVIDIA Entitlement Certificate attached, and asks anyone who does not manage software access or licensing to forward both to the IT administrator. The account created from it opens NGC and the NVIDIA Enterprise Support Portal through the NVIDIA Application Hub.

1. Complete the Spark’s first boot and initial configuration, which the quick start guide asks for before it continues.
2. Forward the order confirmation and the Entitlement Certificate to the administrator who will own the company’s NVIDIA account.
3. Follow the Register link in the certificate’s instructions, complete the registration form and submit it.
4. Open the e-mail that follows, click Log In, enter your e-mail address and click Sign In.
5. Set a password on the Create Your Account page, click Create Account, then click Verify Email Address in the verification e-mail.
6. Log in to the NVIDIA NGC Catalog, set the DGX Spark filter on the search page or open the DGX Spark Collection, then open a software asset to download it.
7. Generate an NGC Personal key with the NGC Catalog service included, then log in with docker login nvcr.io, the username $oauthtoken and the key as the password.

Steps 1 to 6 from NVIDIA’s AI Enterprise for DGX Spark quick start guide (10 September 2026); step 7 as in our activation guide, from NVIDIA’s NGC documentation.

The quick start guide has no serial-number step. Without a certificate that lists the Spark edition, the guide says “do not use the generic NVIDIA AI Enterprise GPU serial-number activation workflow to resolve NIM pull failures” and to authenticate to nvcr.io with an NGC Personal API key instead. Where a certificate asks you to activate a licence and the activation returns an “on hold” or “already claimed” message, the guide says to contact NVIDIA Enterprise Support with the activation reference ID. The guide has no licence server step, and NVIDIA’s support page asks you to confirm that the certificate lists the Spark edition before you open an enterprise case.

NGC organisations, service keys and roles are covered in our guide to activating NVIDIA AI Enterprise licences.

What we supply

We supply the NVIDIA DGX Spark Founders Edition and NVIDIA AI Enterprise licences, the DGX Spark edition among them, on one EU contract and invoice with manufacturer warranty on the hardware. If you need the licence, it goes in the quote alongside the Spark systems, with its term stated. For GPU servers built to order, the per-GPU licences are sized with the cards. Tell us the use case and the team size, and we recommend the edition and term and flag what you do not need.

FAQ