NVIDIA DGX Platform
NVIDIA DGX is NVIDIA's own line of AI systems, from the DGX Spark desktop to rack-scale DGX SuperPOD clusters, delivered together with NVIDIA operations software, reference architectures and support, so an organization can build AI infrastructure on one validated stack.1
Also known as NVIDIA DGX, DGX SuperPOD, DGX BasePOD, DGX Spark, DGX Station, DGX Quantum
At a glance
- What is it?
- The DGX platform is NVIDIA's bundle of AI hardware, software and expert services. On the hardware side it covers data center systems such as DGX B200, B300, GB200, GB300, Rubin NVL8 and Vera Rubin NVL72, plus two desktop machines: DGX Spark (GB10 Superchip) and DGX Station (GB300 Desktop Superchip). DGX SuperPOD is the full-stack cluster design built from those systems, and DGX BasePOD is a reference architecture delivered with storage partners. Software includes NVIDIA Mission Control, Base Command Manager and a qualified operating system stack. NVIDIA's pages differ on DGX Cloud: the DGX platform page describes it as NVIDIA's internal cloud, while the DGX SuperPOD page still presents it as a serverless AI training service for enterprises.12
- What does it do?
- DGX gives an organization dedicated compute for training, fine-tuning and inference, packaged so that compute, storage, networking and management software are tested together instead of assembled piece by piece. A DGX SuperPOD can be configured with any DGX system and, according to NVIDIA, scales to tens of thousands of GPUs. The platform ships with cluster and workload management, an AI-tuned operating system and access to NVIDIA specialists, and it can be run on premises, in a colocation facility, through a managed service provider or in a private cloud. At the small end, DGX Spark lets a developer run and fine-tune models locally before moving work to a larger cluster.12
- Who needs it?
- Enterprises, research centers and public bodies that want to own or colocate dedicated AI infrastructure with a single support path from NVIDIA. Typical buyers are platform teams building an internal AI factory or center of excellence for many users, and organizations that need to keep data and models on their own infrastructure. The desktop systems target individual AI developers, researchers and students who need large local memory for model work.12
- What does it need?23
- A site with power and cooling for the chosen systems (GB200 and GB300 DGX systems are liquid-cooled)
- For GB200/GB300 NVL72 systems: NVIDIA Mission Control for installation and management
- Certified partner storage when building a DGX SuperPOD or DGX BasePOD
- Linux and cluster administration skills (BaseOS/DGX OS, Base Command Manager)
- A purchasing path through NVIDIA or an NVIDIA partner
- What it is not
- DGX is hardware, software and services that you buy or host through partners. For cloud access, NVIDIA's pages are not consistent: the DGX platform page describes DGX Cloud as NVIDIA's internal environment for its own AI work, while the DGX SuperPOD page still offers DGX Cloud as a serverless training service. DGX is also not the only way to buy NVIDIA data center GPUs, because NVIDIA lists HGX systems and NVIDIA-Certified partner servers separately. NVIDIA positions DGX Spark for developing, testing and deploying local agents; in our view it does not replace a production cluster. Buying DGX does not remove facility work: the GB200 and GB300 systems are liquid-cooled rack-scale designs that need matching power and cooling.12
Availability and licensing. DGX systems are sold through NVIDIA and partners. NVIDIA states DGX Spark is shipping; a 64 GB configuration is offered only through participating OEM partners. The product page still shows a notify-me signup, and an NVIDIA blog of October 2, 2026 says it goes on sale October 23, 2026 from Acer, ASUS, Dell, Gigabyte, HP and MSI, from $4,999. The DGX platform page describes DGX Cloud as NVIDIA's internal cloud environment, while the DGX SuperPOD page still presents it as a serverless training service.1245
The problem it solves
Building AI infrastructure from separately sourced servers, network switches, storage and software means an internal team has to integrate, validate and tune every layer before users can run a single job. Problems found after installation are expensive because the hardware is already paid for.
DGX addresses this by selling the compute, the cluster design (DGX SuperPOD or DGX BasePOD), the management software and NVIDIA support as one platform that NVIDIA has already tested together. The trade is less component choice in exchange for a single validated design and one vendor to call when something fails.2
How it works
The platform is organized in layers.
- Systems: DGX servers and rack-scale systems built on Blackwell, Blackwell Ultra and Rubin GPUs, plus the DGX Spark and DGX Station desktops.
- Cluster design: DGX SuperPOD combines those systems with certified partner storage, networking and management into a turnkey cluster; DGX BasePOD is a reference architecture that storage partners deliver.
- Software: BaseOS, which ships as part of DGX OS, supplies drivers, platform settings and diagnostics, and supports Ubuntu, RHEL and Rocky Linux. NVIDIA Mission Control and Base Command Manager handle provisioning, scheduling and monitoring. NVIDIA AI Enterprise supplies the AI software layer.
- Services: NVIDIA Enterprise Services, DGX-Ready colocation data centers and DGX-Ready managed service providers cover deployment and day-to-day operation.
For GB200 and GB300 NVL72 systems, NVIDIA's documentation states that installation and management go through Mission Control rather than the DGX B200 and B300 procedures.136
Diagram as a list
Inference & runtime software
- NVIDIA AI EnterpriseAI frameworks, models and tools with enterprise supportConnects to Mission Control and Base Command Manager
Operations & orchestration
- BaseOS / DGX OSQualified OS, drivers and diagnostics on each systemConnects to DGX systems (B200, B300, GB200, GB300, Rubin NVL8, Vera Rubin NVL72)
- Mission Control and Base Command ManagerProvisioning, scheduling, monitoring and recoveryConnects to DGX SuperPOD / DGX BasePOD, BaseOS / DGX OS
Accelerated computing
- DGX systems (B200, B300, GB200, GB300, Rubin NVL8, Vera Rubin NVL72)GPU compute nodes and rack-scale systemsConnects to DGX SuperPOD / DGX BasePOD
- DGX Spark and DGX StationLocal development systemsConnects to DGX systems (B200, B300, GB200, GB300, Rubin NVL8, Vera Rubin NVL72)
- DGX SuperPOD / DGX BasePODValidated cluster design that groups systems, storage and networkingConnects to Certified partner storage, Cluster networking (InfiniBand and Ethernet)
- Certified partner storageData storage certified by NVIDIA for DGX clusters
Networking, power & facilities
- Cluster networking (InfiniBand and Ethernet)Connects systems and storage inside the PODConnects to DGX systems (B200, B300, GB200, GB300, Rubin NVL8, Vera Rubin NVL72)
Programs & resources
- NVIDIA Enterprise Services and DGX-Ready partnersSupport, training, colocation and managed operationsConnects to DGX SuperPOD / DGX BasePOD
Capabilities
DGX SuperPOD cluster design2
A turnkey cluster that combines DGX systems with storage, networking, software and infrastructure management that NVIDIA has tested together.
Why it matters: Shortens the integration work between buying hardware and giving users a working cluster.
Limits: A large capital project; configuration options follow NVIDIA's validated designs.
DGX BasePOD reference architecture1
A reference architecture for AI infrastructure that NVIDIA says is delivered by storage partners.
Why it matters: Gives smaller deployments a documented design without the full SuperPOD scope.
Limits: A reference design, so the chosen storage partner's components still have to be integrated and supported.
Desktop AI systems14
DGX Spark uses the GB10 Superchip with up to 128 GB of unified memory and can link up to four units through ConnectX networking; DGX Station uses the GB300 Desktop Superchip with 748 GB of coherent memory.
Why it matters: Lets developers work with large models locally before moving to a cluster.
Limits: NVIDIA positions these desktops for developing, testing and deploying local agents; we would not plan on them for serving production traffic at data center scale.
Qualified operating system stack6
BaseOS, part of DGX OS, adds drivers, platform-specific settings and diagnostic and monitoring tools, with support for Ubuntu, RHEL and Rocky Linux.
Why it matters: Reduces driver and firmware mismatch problems on new systems.
Limits: Limited to the listed Linux distributions.
Integrated operations software13
DGX includes NVIDIA Mission Control for AI factory operations, and NVIDIA AI Enterprise is optimized for the platform and used together with Base Command Manager.
Why it matters: Provisioning, scheduling and health monitoring come from the same vendor as the hardware.
Limits: Licensing steps differ by system; for GB200/GB300 NVL72, NVIDIA's Mission Control documentation says NVIDIA Installer Services generate and provision the required licenses.
Flexible hosting options1
DGX can run on premises, in DGX-Ready colocation data centers, through DGX-Ready managed service providers or in private clouds.
Why it matters: Organizations without suitable facilities can still own DGX capacity.
Limits: Colocation and managed services are partner offerings with their own contracts.
Practical use cases
Many teams across a large organization each buy or rent their own GPUs, which fragments tools, data access and security controls.
- Approach
- Centralize compute, MLOps tools and practices in one internal AI factory built on DGX SuperPOD, shared by many users.
- Role of NVIDIA DGX Platform
- DGX SuperPOD provides the shared cluster; Mission Control and AI Enterprise provide management and software.
- Data, infrastructure and skills
- Executive sponsorship for a shared platform, facility capacity, a platform team.
- Type of benefit
- Shared infrastructure with central governance
- Caveats
- A central platform needs internal chargeback or quota rules so one group does not monopolize capacity.
- First step
- Inventory current GPU spend and users across teams to size a shared cluster.
Sources 1
An organization must build language models for a specific language or market while keeping data and models under its own control.
- Approach
- Train and serve the models on owned DGX SuperPOD infrastructure rather than on shared public cloud capacity.
- Role of NVIDIA DGX Platform
- DGX SuperPOD supplies dedicated training and inference capacity.
- Data, infrastructure and skills
- Training data, model engineering staff, long-term capacity planning.
- Type of benefit
- Data and model sovereignty
- Caveats
- Owning hardware shifts utilization risk to the buyer.
- First step
- Define the model sizes and training schedule to estimate the system count.
Sources 2
Developers need to test agents and large models locally without sending data to a cloud endpoint.
- Approach
- Use DGX Spark as a desktop development system with the NVIDIA AI software stack preinstalled, then move finished work to a data center or cloud.
- Role of NVIDIA DGX Platform
- DGX Spark runs local inference and fine-tuning.
- Data, infrastructure and skills
- Desk space and network access; familiarity with the NVIDIA software stack.
- Type of benefit
- Local development with data kept on site
- Caveats
- NVIDIA states the 128 GB DGX Spark handles inference on models up to about 200 billion parameters and fine-tuning up to 70 billion, and the 64 GB model supports models up to 100 billion; larger work needs a cluster.
- First step
- List the models and agent frameworks you plan to test and check them against DGX Spark memory limits.
Who uses it
Danish Centre for AI Innovation (DCAI) · Research and AI infrastructure (sovereign AI)
DCAI Gefion: Denmark's sovereign AI supercomputer on DGX SuperPOD
Gefion, operated by the Danish Centre for AI Innovation, is an NVIDIA DGX SuperPOD with 1,528 H100 GPUs, funded by the Novo Nordisk Foundation and Denmark's export and investment fund. It went live in October 2024, placed 21st on the November 2024 TOP500 list, and now serves researchers and companies.
In production
Deutsche Telekom (T-Systems) · Telecommunications and cloud services
Deutsche Telekom Industrial AI Cloud: an NVIDIA-based AI factory in Munich
Deutsche Telekom opened its Industrial AI Cloud in Munich on 4 February 2026, with nearly 10,000 NVIDIA Blackwell GPUs in DGX B200 systems and RTX PRO Servers. Telekom says it was over a third utilized at opening, with Agile Robots and PhysicsX among early users.
In production
Mayo Clinic · Healthcare and medical research
Mayo Clinic: a DGX SuperPOD with DGX B200 for pathology foundation models
Mayo Clinic deployed an NVIDIA DGX SuperPOD with DGX B200 systems in July 2025 and says its first work will be building foundation models for pathology, drug discovery and precision medicine. Mayo states the system cuts four weeks of slide analysis and model work to one; no measurement is published.
In production
BMW Group · Automotive manufacturing
BMW Group: planning car plants in an Omniverse-based Virtual Factory
BMW Group plans and checks its car plants in a Virtual Factory built on NVIDIA Omniverse and OpenUSD. BMW says digital twins now cover more than 30 production sites, cut collision checks for new models from almost four weeks to about three days, and are projected to lower planning costs by up to 30 percent.
Scaling
Works with
Optional integration
- NVIDIA Mission ControlMission Control operates DGX clusters and is required for GB200/GB300 NVL72 installation.
- NVIDIA AI EnterpriseNVIDIA states AI Enterprise is optimized for the DGX platform.
Complementary tools
- NVIDIA Mission ControlRecommended software for DGX B200/B300 and required for GB200/GB300 NVL72 installation.
- NVIDIA Nsight Developer ToolsNsight Systems profiles workloads on DGX systems, including multi-node runs.
- NVIDIA Run:aiRun:ai, included in Mission Control, schedules workloads on DGX GPU pools.
Optional integration for
- NVIDIA Holoscan SDKDGX Spark is a supported platform.
- NVIDIA OpenShellNVIDIA states OpenShell runs on DGX Spark and DGX Station.
- NVIDIA DeepStream SDKThe documentation includes a DGX Spark setup; on DGX Spark DeepStream runs from its container.
Relationship labels follow NVIDIA's documentation. "Alternative approaches" does not mean one is better: each profile says when it fits.
Getting started
Pick the system class
Match the workload to DGX Spark or DGX Station (individual development), DGX BasePOD (reference architecture with a storage partner) or DGX SuperPOD (turnkey cluster).
Check: A written sizing note that names the system type and the number of users or jobs it must serve.
Decide where it will run
Choose between your own data center, a DGX-Ready colocation data center or a DGX-Ready managed service provider.
Check: The facility confirms it can supply the power and liquid cooling the chosen systems need.
Find the right installation path
Use the DGX documentation hub for your system. For GB200 or GB300 NVL72 systems, follow the Mission Control installation guides, not the DGX B200/B300 procedures.
Check: The installation guide and release notes for your exact system and software version are identified.
Plan the software layer3
Decide on BaseOS or DGX OS, Base Command Manager or Mission Control, and whether to license NVIDIA AI Enterprise.
Check: For GB200/GB300 NVL72, NVIDIA Installer Services have provisioned the required licenses; for other systems, the license plan is confirmed with NVIDIA or the reseller before installation day.
Engage NVIDIA or a partner
Use the Get DGX path on the product page to reach NVIDIA or a partner for configuration and quoting.
Check: A configuration proposal that lists systems, storage, networking and support terms.
Official resources
Could this technology help you?
Describe your project to the Solution Architect. It starts with NVIDIA DGX Platform as context but recommends independently, including when you do not need it.
Sources
Each statement above links to the source it comes from. Labels say who reported it.
- NVIDIA DGX Platform (opens in a new tab)
- NVIDIA DGX SuperPOD (opens in a new tab)
- NVIDIA Mission Control documentation hub (opens in a new tab)
- NVIDIA DGX Spark product page (opens in a new tab)
- NVIDIA blog: DGX Spark 64GB (opens in a new tab)
- NVIDIA DGX Platform documentation hub (opens in a new tab)
Thank you. Your correction was sent.
The editors check it against the sources. If you left an email address, they may reply about it.