NVIDIA Omniverse Enterprise licensing in 2026: free use, paid support and GPU requirements

Omniverse Enterprise licence and price in 2026

NVIDIA Omniverse has needed no paid licence since May 2026. NVIDIA's Omniverse licensing page, read 12 November 2026, states: "As of May 2026, Omniverse is freely available for both development and production use with no NVIDIA AI Enterprise subscription required." What a company can still buy is enterprise support, as an NVIDIA AI Enterprise subscription counted per GPU; "Omniverse Enterprise" is a legacy name in NVIDIA's product terms, and NVIDIA publishes no price for Omniverse.

The same page answers the questions behind a search for an "Omniverse Enterprise licence".

NVIDIA, "Omniverse Licensing" in the Omniverse developer documentation, read 12 November 2026.

The terms govern the use of Omniverse "for both individuals and commercial customers", so a company works under the same terms as a single developer. According to the page, end customers buy the NVIDIA AI Enterprise subscription through NVIDIA's sales channel or in a cloud marketplace, and software makers that embed or redistribute Omniverse in a product of their own are offered an embedded licence with Enterprise Support. One condition comes from the product-specific terms for NVIDIA AI products: Omniverse is among the components "licensed solely to run on systems with NVIDIA Platforms".

NVIDIA's Software Licensing Overview and FAQ of 11 September 2026 files Omniverse accordingly. Its free Community Products "can be used for both development and production and receive community-level support", and the examples it gives are "NVIDIA Triton containers, NVIDIA NeMo containers, and NVIDIA Omniverse". Which other NVIDIA tools are free for commercial use is the subject of our article CUDA, TensorRT, NIM: free or NVIDIA AI Enterprise.

What became of the name Omniverse Enterprise

NVIDIA's product-specific terms, last modified on 15 April 2026, retire the name. Certain legacy definitions, including "Omniverse Enterprise", "have been integrated or rebranded within this unified framework under 'NVIDIA AI Enterprise'". Two sentences for existing customers follow: "Your current Subscription and Part Number remain valid until your Subscription Term ends." and "New purchases or renewals will transition to updated Part Numbers."

A company that holds an Omniverse Enterprise subscription therefore keeps it until the end of its term, and a renewal moves to an updated part number. A tender, a budget line or an older quote that still says "Omniverse Enterprise licence" translates into two statements today: Omniverse itself is free, and an NVIDIA AI Enterprise subscription per GPU is needed where enterprise support is required.

NVIDIA's own pages still carry the old name in places. The Omniverse releases page speaks of enterprise support for Omniverse Enterprise subscribers, and a cloud footnote on the NVIDIA AI Enterprise pricing page notes that the Omniverse Enterprise Workstation has been renamed the Omniverse Production Workstation. The address of the former Omniverse Enterprise page now serves the same content as NVIDIA's general Omniverse page. For the licence question we rely on the licensing page and the product terms.

What NVIDIA AI Enterprise adds to Omniverse

NVIDIA's licensing guide for NVIDIA AI Enterprise, last updated on 2 September 2026, describes NVIDIA Omniverse, "included with NVIDIA AI Enterprise", as a collection of libraries and microservices for industrial digitalisation workflows, and the definition of NVIDIA AI Enterprise in the product terms names "NVIDIA Omniverse (including Kit)". Since the software itself is free, the subscription pays for support.

NVIDIA "Omniverse Licensing" page; NVIDIA Enterprise Licensing Guide (last updated 2 September 2026); Product-Specific Terms for NVIDIA AI Products (last modified 15 April 2026); all read 12 November 2026.

Under Business Standard, NVIDIA's support guide gives technical support in local business hours and an initial response to a severity 1 case within four business hours. The count follows the hardware. The licensing guide requires a licence for every GPU installed in a server or workstation that hosts NVIDIA AI Enterprise software, so a workstation with one card needs one licence and a render server with eight cards needs eight. NVIDIA supports the software only on NVIDIA-Certified Systems; what that label covers is explained in NVIDIA-Certified Systems.

On price, NVIDIA's pricing page, last updated on 2 September 2026, prices NVIDIA AI Enterprise per GPU; a separate price for Omniverse is not published. Forms, terms and counting of the subscription are set out in our guide to NVIDIA AI Enterprise licensing.

Three details matter when the support is planned. Omniverse is released in two branches: a Production Branch every six months with API consistency for nine months, and a monthly Feature Branch that is supported until the next one and that NVIDIA recommends for development, testing and review only. Production Branch 26h1 was released in May 2026, and NVIDIA names NGC as the primary location for enterprise-supported Omniverse content.

Not everything is covered. The libraries ovrtx, ovphysx, ovstream, ovstorage, ovui and ovpackage are marked as pre-release software that is not enterprise-supported, and Launcher, USD Composer and USD Explorer are listed as legacy and no longer supported.

For the cards, NVIDIA documents no included subscription. Neither the licensing guide nor the pricing page, the activation page of 7 August 2026 or the RTX PRO Server page of 11 June 2026 states that NVIDIA AI Enterprise is included with RTX PRO cards or RTX PRO Servers. The product terms allow NVIDIA to offer subscriptions at different prices and terms solely for designated platforms such as the RTX PRO Server, and such a licence is restricted to that platform. For a first test, NVIDIA's 90-day trial of NVIDIA AI Enterprise includes Omniverse.

We supply NVIDIA AI Enterprise, the subscription through which NVIDIA provides enterprise support for Omniverse; name the GPU models, their number and the term you want, and a written quote follows within one business day.

Omniverse GPU requirements as NVIDIA publishes them

NVIDIA's technical requirements for Omniverse, in the page version of 9 November 2026, and its Isaac Sim requirements of 14 October 2026 name cards, and only some tiers carry a memory figure.

NVIDIA Omniverse developer documentation, technical requirements (9 November 2026); Isaac Sim requirements (14 October 2026); read 12 November 2026.

Processors, system memory, storage and streaming for each tier are laid out in our article on Omniverse system requirements, and robot simulation in GPU server for Isaac Sim and Isaac Lab. Three further statements decide the choice of card.

The first is RTX. For Omniverse SDKs on non-RTX GPUs NVIDIA gives no support commitment, and the Isaac Sim page states that "GPUs without RT Cores (A100, H100) are not supported."

The second is the feature table by GPU architecture. Its rows for Blackwell RTX PRO cards, compute capability 12.0, and for Ada Lovelace cards, 8.9, mark every listed RTX feature as supported. The B200 and B300 data-centre GPUs, 10.0, lack DLSS Frame Generation, Shader Execution Reordering, Opacity Micro-Map and Motion BVH, and Hopper GPUs lack DLSS Ray Reconstruction as well. An AI server with H200 NVL cards is therefore not the place for Omniverse rendering, although each H200 NVL includes a five-year NVIDIA AI Enterprise subscription.

The third is the driver. NVIDIA lists 580.95.05 (R580) and 595.58.03 (R595) as validated on Linux and 581.42 and 595.97 on Windows, and the list has no R620 entry. NVIDIA's current RTX Enterprise drivers for Windows are 621.14 on R620, of 4 November 2026, 597.30 and 582.99, each newer than the validated version of its branch. NVIDIA's note on that reads: "Newer drivers may function properly, but have not been fully tested for stability and compatibility."

Some things the requirements page leaves out: it has no list of certified workstations and says nothing about vGPU or MIG. NVIDIA's OVX page of 4 January 2026 lists configurations with four and eight L40S cards, names no RTX PRO card and makes no licence statement. The RTX PRO Server page names Omniverse and NVIDIA AI Enterprise as workloads and says that RTX PRO Servers are NVIDIA-Certified.

The cards we supply against NVIDIA's figures

memory from NVIDIA's datasheets; compute capability from NVIDIA's "CUDA GPU Compute Capability" page (16 October 2026), where the Server Edition stands in the data-centre column; feature rows from the Omniverse technical requirements (9 November 2026); Isaac Sim figures as in the table above.

The table compares memory and the feature row and nothing else. NVIDIA's requirement pages name none of these eleven cards: their tiers name the GeForce RTX 3070, RTX 4080 and RTX 5080 and the RTX PRO 6000 Blackwell. A row shows how the memory of a card sits against Isaac Sim's figures. It does not make the card one that NVIDIA lists for a tier. "Not assigned" means that NVIDIA's CUDA page does not list the card, so we attach it to no row of the feature table.

Which card for which Omniverse task

The following is our reading of NVIDIA's figures; NVIDIA gives no sizing for these cards.

The workstation as NVIDIA specifies it has an RTX PRO 6000 Blackwell with 96 GB. We supply that card in its Workstation, Max-Q and Server editions, described on our RTX PRO 6000 page, and RTX PRO 6000 alternatives: 48 GB and 32 GB cards shows which jobs the smaller cards take over.

Among the eleven cards above, the NVIDIA RTX PRO 5000 Blackwell with 48 GB is the one whose memory equals the figure of Isaac Sim's ideal tier. It draws 300 W and takes two slots.

The 24 GB and 32 GB cards lie between Isaac Sim's two memory figures: the NVIDIA RTX PRO 4500 Blackwell Workstation Edition with 32 GB at 200 W in two slots, the NVIDIA RTX PRO 4000 Blackwell with 24 GB at 145 W in one slot and, on the Ada side, the NVIDIA RTX 5000 Ada Generation with 32 GB and the NVIDIA RTX 4500 Ada Generation with 24 GB. How far they carry depends on the scene and the sensors, which NVIDIA does not express in GB. The systems around these cards are the subject of GPU workstation builds from RTX PRO 2000 to 5000, and the 32 GB card in a real-time engine of RTX PRO 4500 for Unreal Engine and virtual production.

The three 16 GB cards, the NVIDIA RTX PRO 2000 Blackwell, the NVIDIA RTX 2000 Ada Generation and the NVIDIA RTX 2000E Ada Generation, stand exactly at Isaac Sim's minimum memory figure, with no margin above it. The NVIDIA RTX A1000 with 8 GB and the NVIDIA RTX A400 with 4 GB lie below it, and we do not recommend them for Omniverse.

In a server, the NVIDIA RTX PRO 4500 Blackwell Server Edition is the passive single-slot card with 32 GB at 165 W, on NVIDIA's vGPU list from release 20.0. NVIDIA's own server specification names eight RTX PRO 6000 Blackwell Server Edition cards. We build servers to order with either card and supply Supermicro, Gigabyte, ASUS, Dell and HPE servers per project; where NVIDIA support for NVIDIA AI Enterprise is planned, the server model has to be on NVIDIA's list of certified systems. For hosts that run under VMware, NVIDIA AI Enterprise on vSphere covers licence and vGPU.

We supply these cards across the EU, as cards or in workstations and servers built to order; tell us the application, the size of your scenes and the number of seats, and we quote the cards with or without NVIDIA AI Enterprise.

What to check before renewing or ordering support

Where an Omniverse Enterprise subscription is still running, or support is to be ordered for the first time, four checks prepare the order:

1. The end date and the part number on the entitlement. The current subscription stays valid until that date, and a renewal carries an updated part number.
2. The number of GPUs in every server and workstation that will run the supported software. The subscription is counted per installed GPU.
3. The system models. NVIDIA supports NVIDIA AI Enterprise only on NVIDIA-Certified Systems.
4. The release branch in use. NVIDIA recommends feature branches for development, testing and review only, so a supported production system belongs on a Production Branch.

For a quote, add the application and the size of the scenes to these four points; the list can also go by e-mail to request@eurokommerz.at.

FAQ