NVIDIA DGX Station available now for purchase
The NVIDIA® DGX Station has been launched, and is now available for order from Pi3g.
The DGX Station is a powerful AI supercomputer powered by the NVIDIA® GB300 Grace Blackwell Ultra Desktop Superchip.
pi3g is a value added distributor for AI hardware solutions based in Leipzig, Germany. Contact us for offers, and to determine if the DGX Station is the right fit for you.
What is the specification (specs) for the DGX Station?
- NVIDIA® GB300 Grace Blackwell Superchip with up to 20 petaFLOPS of AI performance
- 748GB coherent memory (496GB LPDDR5x @ 396 GB/s + 252GB HBM3e @ 7.1 TB/s)
- NVIDIA ConnectX-8 SuperNIC – up to 800 Gbit/s – supports linking up to two DGX Stations together.
- NVIDIA Multi-Instance GPU (MIG) Technology, allows to partition the GPU in up to seven isolated instances for independent work loads
- Processor: 1x NVIDIA Grace ™ 72-core Neoverse V2 CPU (ARM-based)
- 10GBase-T Ethernet port with Marvell AQC113
- Four NVMe M.2 slots for storage
- 3 x PCIe 5.0 x16 slots
- 1 x PCIe 2.0 2230 M.2 for WiFi / BT
- 6 x USB 3 Type A ports
- 2 x USB Type C (USB3.1)
- 1600W power supply
The NVIDIA DGX Station features NVIDIA Blackwell Generation Tensor Cores, enabling 4-bit floating point (NVFP4). This allows you to use models with more parameters (up to 1 Trillion parameter model class), while still maintaining high accuracy.
Form factor: tower, about 250 mm W x 530 mm H x 590 mm D (depends on vendor)
How does the DGX Station compare performance-wise?
The following is an overview of different NVIDIA-based solutions:
- (Server) NVIDIA HGX H200 SXM (8-GPU) – 32 petaFLOPS
- (Supercomputer) NVIDIA DGX Station GB300 – 20 petaFLOPS
- (Server) NVIDIA HGX H200 SXM (4-GPU) – 16 petaFLOPS
- (High-end workstation) GPU Workstation with 4x RTX PRO 6000 Blackwell WK ED – 8 petaFLOPS
- (Supercomputer) DGX Spark – 1 petaFLOPS
Please note – the petaFLOPS are estimates based on FP4. Model performance may vary with different model sizes.
The screenshot, which our first customer provided to us, shows the performance of DeepSeek-V4-Flash-NVFP4 and Qwen2.5-Coder-32B-Instruct on a DGX Station GB300.
What are some possible applications for the DGX Station?
DGX Station systems are enterprise grade systems, designed for local on-premise office use (as compared to the server systems from NVIDIA).
As workstations they can be used by individual researchers, or using the networking and the NVIDIA MIG Technology shared by several individuals in the enterprise. If they are configured as local AI inference servers, they can serve tens to hundreds of users concurrently (depending on the model).
The memory bandwidth on the HBM3e memory portion (252GB) is a massive 7.1 TB/s. This is faster than the H200 with 4.8 TB/s of memory bandwidth. This enables key applications:
- Running large language models – especially MoE (mixture of experts) models
- Running and fine-tuning frontier AI models of up to 1 Trillion parameters
- Support hundreds of parallel AI agents running simultaneously on the same platform
- Massive RAG databases for AI-enhanced document workflows – ingest all of the company’s documents locally
- BioNeMo – AI-driven biology
A huge difference from previous AI hardware available for office purposes is the much improved suitability for AI development. It is the first mainstream, commercial product to bring HBM3e (High bandwidth memory) directly into a standard office or laboratory setting.
The DGX Station has a preconfigured AI software stack based on Ubuntu with NVIDIA AI Developer tools. It includes the necessary CUDA-X libraries. We can customize the system even further for you, and develop a full turn-key AI solution based on the DGX Station.
The following industries will benefit from using the NVIDIA DGX Station:
- Healthcare & Medical Research – Medical Image analysis, Automated Disease analysis at scale
- Enterprise (AI Agents & Copilot Applications) – Data Analysis and Insights, Automation of repetitive tasks
- Smart Cities & Edge AI – Smart Energy Grid, Smart Security, Smart Transportation
- Scientific Research – Data Analysis and Pattern Recognition at scale, Model building and simulation
- Automotive industry – Computational Fluid dynamics, Aerodynamic optimization, Fault prediction and diagnosis, Material design
- Aerospace Industry – Aircraft design and optimization
- Pharma & Chemical Industries – Drug Design, Drug Property Prediction, Chemical Research
- Oil & Gas – Seismic Data Analysis, Drilling Optimization
The following stakeholders will benefit from using the DGX Station:
- Software developers and security analysts
- Data analysts and family offices, and other stake holders which need to process a lot of data quickly
- AI Agent
- AI Trainer
- AI Researcher
- AI Developer
- Data Scientist
- Machine Learning Application Developer
- AI Startup Founder
- Customer support teams (for local customer support acceleration + RAG databases)
- Any other team wanting to run local inference workloads
Multiple NVIDIA DGX Stations can be linked using the ConnectX-8 SuperNICs for increased performance. This allows you to scale model capacity and performance even further, or increase the precision of the model you are currently running.
The DGX Station allows you to install NVIDIA RTX PRO 6000 / 4000 / 2000 GPUs to enhance the massive AI compute capability with powerful rendering and simulation workload capabilities.
What is a key application and reason for the purchase of the DGX Station?
The key reason for purchasing a DGX Station is reducing risk and dependency on cloud providers.
As online AI platforms are becoming increasingly more regulated by the respective governments, this may cause disruptions to teams building on AI workflows relying on these cloud providers.
A recent example is Anthropic suspending access to Fable 5 and Mythic 5 for all customers.
AI which you install and run locally on a DGX Station is under your control. An additional benefit is full privacy for your corporate data.
The DGX Station comes with key enterprise features for easy integration in your corporate IT landscape:
IT manageability and enterprise features
- out-of-band telemetry via BMC
- Redfish API support
- NVIDIA Data Center GPU Manager
- visibility, monitoring, remote management capabilities
- hardware root of trust
- enterprise secure boot
Do I need to worry about noise or special electrical sockets for the DGX Station?
The NVIDIA DGX Station uses a closed-loop liquid cooling system designed specifically for office-friendly acoustics.
Since it operates on a single 1600 Watt power supply, it can be run on standard office power circuits without custom electrical overhauls.
You don’t need to accomodate data-center grade hardware in your office anymore, the DGX Station brings data-center level performance directly to you.
What are some OEM names for the DGX Station?
The NVIDIA® DGX Station technology is available from different vendors – it has the same core specification in each case, with different cases and some other changes, for example modified cooling systems.
Due to very strict NVIDIA checks, the systems should all perform comparably.
- MSI XpertStation WS300 (Model WS300T60L)
- Dell Pro Max With GB300
- HP ZGX Fury AI Station (HP ZGX Fury G1n AI Station)
- ASUS ExpertCenter Pro ET900N G3 // P/N 90WS00B2-M00020 // EAN 4711636596701 // UPC 199291596702
- Supermicro Super AI Station GB300 (Model ARS-511GD-NB-LCC )
- Exxact Valence VWS-158270643 and Exxact Valence VWS-11703208
- Gigabyte W775-V10-L01
We will add reviews for the individual systems here. In the meantime, we have collected a lot of information already – if you are comparing different DGX Station systems, please reach out to us for a free and vendor-neutral consultation.
How much does the DGX Station cost in 2026?
As of August 2026, the DGX Station costs:
Indicative price in USD: 105.000 USD excluding VAT and shipment.
Indicative price in EUR: 95.000 EUR excluding VAT and shipment
Please note: prices are expected to rise, due to the current memory market situation. See below for more detailed pricing information by DGX station brand.
If you order three or more DGX Stations, we will ship them for free within Europe.
If you order ten or more DGX Stations, we will ship them for free worldwide.
The pricing excludes SSD storage and set-up services.
We also offer value-added services – we can customize the DGX Station for you, add SSD storage, an additional GPU, pre-install applications, or deliver bespoke turn-key solutions based on the DGX Station, for example RAGs, agentic AI applications, AI document workflows, and other AI solutions.
What is the lead time of the DGX Station in August 2026?
Here is a break down of the lead time and pricing of some of the DGX Station systems, by vendor. All prices excluding shipping and taxes. Our prices are indicative. The prices will be finalized, once your purchase with us is approved by NVIDIA.
The price is likely to increase significantly (10 – 20 % or more) in the next weeks, due to memory price increases. The Euro pricing is dependent on the USD exchange rate.
If you need customization and bespoke installations, the lead time is extended by about 1-2 weeks from our side.
As you can see below, the lead time of the DGX station in Agust 2026 is from 6-13 weeks, depending on the vendor. Price for the DGX station varies from $103.5k to $123k. The systems vary in their installed storage, add-on GPU, and on-site service. Get in touch with us to help you choose the optimal DGX Station for you.
DGX Station lead times & prices last updated on 14.8.2026
| DGX Station System | Storage / extra GPU | price USD (excluding VAT) | price EUR (excluding VAT) | lead time / estimated arrival |
| Supermicro Super AI Station GB300 (1× NVIDIA Grace + Blackwell) | 2x 1TB M.2 + 2x 2TB M.2 | $103.500 | 93.000 € | 8-9 weeks, October |
| MSI XpertStation WS300 GB300 Workstation (Grace + Blackwell, 2× M.2 NVMe OS) | 2x 2TB M.2 | $103.500 | 93.000 € | 10-12 weeks, October-November |
| Exxact Valence Workstation VWS-158270643 – NVIDIA GB300 Tower Workstation (based on MSI) | 2x 2TB M.2 RTX Pro 2000 BW | $105.000 | 95.000 € | 6-8 weeks, October |
| Gigabyte W775-V10-L01 Workstation | 4x 2TB M.2 | $112.000 | 99.500 € | 8-10 weeks, October |
| HP ZGX Fury AI Station | 2 x 2TB M.2 + 1 x 4TB M.2 | $122.500 | 110.000 € | EU: October USA: September |
| ASUS ExpertCenter Pro ET900N G3 | 4x 2TB M.2 3 years on-site service | $123.000 | 110.500 € | 10-13 weeks, October-November |
| Dell Pro Max With GB300 | TBD | TBD | TBD | TBD |
Available add-on GPU options and expected add-on prices:
| GPU name | GPU RAM | price net USD | price net EUR |
| NVIDIA RTX PRO 6000 Workstation Edition | 96 GB @ 1729 GB/s 4000 AI Tops | $14.000 | 12.500 € |
| RTX PRO 6000 Blackwell Max-Q Workstation Edition | 96 GB @ 1792 GB/s 3511 AI Tops | $14.000 | 12.500 € |
| RTX PRO 4000 Blackwell SFF Edition | 24 GB @ 432 GB/s 770 AI Tops | $2.500 | 2.200 € |
| RTX PRO 2000 Blackwell | 16 GB @ 288 GB/s 545 AI Tops | $1.200 | 1.100 € |
Please note: the secondary GPU is primarily to be used for providing more displays; due to the additional power usage, the system’s AI performance is not expected to be increasable by the additional GPU.
How does the price develop for the NVIDIA DGX Station GB300? What is the expected price in 2027?
We will track the price development for the Supermicro Super AI Station ARS-511GD-NB-LCC-01-G2 DGX Station system here. It is an example for other DGX Stations as well.
| Month | price net USD | price net EUR | note |
| July 2026 | $99.500 | 88.500 € | initial stock availability |
| August 2026 | $103.500 | 93.000 € | price increased (+$4k), USD/EUR trending down |
| expected price 2027 | $145.000 | 125.000 € | further price increases expected in the DRAM market. |
We currently expect the memory to be significantly more expensive in 2027, and the prices to continue to rise.
Are discounts available for the NVIDIA DGX Station?
Currently there are two programs:
Inquire with us to determine if you are eligible. In our blog posts (linked above) we have described the procedure necessary to secure the discount for the DGX Station.
What will be the future lead time of the NVIDIA DGX Station after the initial stock is exhausted?
We got an indication from an OEM that there would be lead times of about 8-10 weeks for individual new orders, once the initial stock of that OEM’s DGX Station would be exhausted. This might be different across OEMs.
Do you sell the DGX Station to universities, schools, and research institutes?
Yes, we sell the DGX Station to universities, schools and research institutes. We will need an official purchase order from your institution. Get in touch with us for details, to discuss the purchase of your NVIDIA® DGX Station.
An educational discount is available, inquire with us for the discount, please.
NVIDIA DGX Station FAQs
We’ve moved other questions and answers for the DGX Station into it’s own blogpost. The DGX station FAQ blogpost is a valuable resource of technical and commercial questions around the DGX station, please read it to get the information you need to make an informed purchase of the NVIDIA DGX Station.
English 
Deutsch (Sie)