DGX Station
Workstation, GB300 Desktop superchip and OEM models
| Indicative price | €88,000 to €110,500 excl. VAT depending on the OEM (Supermicro €88,000, MSI €90,000, Exxact €95,000, Gigabyte €99,500, HP €110,000, ASUS €110,500), excluding VAT and shipping, listed by the reseller pi3g on 14 September 2026; Dell at €173,000 excl. VAT with 16 TB of storage and three years of on-site service, outside the range; the site's calculations use €150,000 excl. VAT (€180,000 incl. VAT), the price used from October 2026 after the observed increase (pi3g reading of 14 September 2026: €95,000 excl. VAT for the Exxact Valence configuration, the previous reference; price guide) |
|---|---|
| Memory | 748 GB coherent: 252 GB HBM3e (7.1 TB/s) and 496 GB LPDDR5X (396 GB/s); 784 GB had been announced in 2025 |
| GPU | Blackwell Ultra GB300 Desktop and 72-core Grace, linked by NVLink-C2C |
| Compute | 20 PFLOPS FP4 with sparsity (15 dense) |
| Network | ConnectX-8 SuperNIC, up to 800 Gb/s |
| Power | 1,600 W, 20 A circuit |
| Models | up to 1,000 billion parameters according to NVIDIA |
| Form factor | Workstation |
What does the DGX Station enable?
The DGX Station lets teams develop, fine-tune and serve models of up to 1,000 billion parameters on premises according to NVIDIA, without a data center. Engineers run fine-tuning with Unsloth and inference with vLLM right at their desk.
Who is the DGX Station designed for?
The DGX Station targets mid-sized companies, laboratories and departments that want to keep sensitive data in house without a server room: a compute rack fits in a desktop chassis. Its 1,600 W still require a 20 A electrical circuit, according to NVIDIA's guide.
Why does the DGX Station price change from month to month?
The DGX Station price follows the memory shortage, which moves prices from one month to the next. At pi3g, the Supermicro Super AI Station went from €88,500 excl. VAT in July 2026 to €93,000 in August, then €88,000 in September. The reseller expects about €125,000 excl. VAT in 2027, with memory set to become even more expensive.
Which OEMs build the DGX Station?
NVIDIA does not sell the DGX Station directly: six OEMs build it around the same GB300 Desktop superchip, and two integrators configure it to order. The models differ mainly in storage, cooling and service, which explains the price gap.
| OEM and model | Configuration listed | Price excl. VAT at pi3g, 14 September 2026 | Other listings |
|---|---|---|---|
| Supermicro Super AI Station (ARS-511GD-NB-LCC) | Liquid cooling, 2 × 1 TB and 2 × 2 TB M.2 | €88,000 | $92,887 excl. VAT according to TechRadar, 8 to 16 week lead time |
| MSI XpertStation WS300 (formerly CT60-S8060) | ATX form factor, 1,600 W power supply, 2 × 2 TB M.2 | €90,000 | $99,999 at Newegg (out of stock), $125,187 at CDW (MSI) |
| Exxact Valence VWS-158270643 (integrator, MSI based) | 2 × 2 TB M.2, RTX PRO 2000 display card | €95,000 | from $96,250 at Exxact |
| Gigabyte W775-V10-L01 | Liquid cooling, 4 × 2 TB M.2 | €99,500 | $116,999 at Newegg (out of stock), Gigabyte page |
| HP ZGX Fury G1n | RHEL certified, Red Hat AI Factory offer, 2 × 2 TB and 1 × 4 TB M.2 | €110,000 | orderable since 9 September 2026 (StorageReview) |
| ASUS ExpertCenter Pro ET900N G3 | 4 × 2 TB M.2, three years of on-site service | €110,500 | $99,999 suggested price, €95,750 excl. VAT at Iodis (ASUS) |
| Dell Pro Max with GB300 (FCT6263) | MaxCool liquid cooling, 4 × 4 TB M.2, RTX PRO 2000, three years of on-site service | €173,000, outside the range | £151,307 excl. VAT in the United Kingdom (Dell) |
| BIZON (integrator) | Custom configurations | not listed | from $112,159 at BIZON |
Dollar prices are those of the cited sources, without conversion. No Lambda DGX Station was found as of 1 October 2026.
DGX Station generations: from V100 to GB300
The DGX Station has gone through three generations. The DGX Station V100 (2017) and the DGX Station A100 (2020) combined four discrete GPUs linked by NVLink, with 128 GB of HBM2 and up to 320 GB of HBM2e respectively. The 2026 generation moves to a single superchip that merges a Blackwell Ultra GPU and a Grace processor in 748 GB of coherent memory.
748 GB unified memory architecture
The 748 GB unified memory architecture combines 252 GB of HBM3e on the GPU side and 496 GB of LPDDR5X on the processor side, linked by NVLink-C2C at 900 GB/s. This coherence avoids copies over the PCIe bus and keeps tensors in a shared address space.
GPUDirect Storage
On platforms equipped with ConnectX cards, GPUDirect Storage sets up a direct path between NVMe storage, local or remote, and GPU memory, bypassing the processor buffer. Compatible arrays such as NetApp, VAST, DDN or WEKA can thus feed training and RAG.
Which software environment for the DGX Station?
The DGX Station ships with DGX OS, an Ubuntu adapted by NVIDIA with CUDA, and accepts an open stack, vLLM, TensorRT-LLM or Unsloth, with no recurring licence. For contractual support, NVIDIA AI Enterprise gives access to certified NIM containers, licensed per GPU.
Frequently asked questions
How much does a DGX Station cost?
From €88,000 excl. VAT (Supermicro) to €110,500 excl. VAT (ASUS) depending on the OEM, excluding VAT and shipping, according to the reseller pi3g on 14 September 2026. The Dell configuration, with 16 TB of storage and three years of on-site service, is listed separately at €173,000 excl. VAT. Since October 2026, the site's calculations use €150,000 excl. VAT, after the observed increase.
What is the DGX Station used for?
To develop, fine-tune and serve large models on premises without a server room: a compute rack in a desktop chassis for a mid-sized company.
Which runtime for the DGX Station?
vLLM to serve large models, Unsloth for fine-tuning.
DGX Station or DGX Spark?
The DGX Spark is a compact workstation for models up to 200 billion parameters; the DGX Station, more powerful and more expensive, targets developing and serving larger models.
How much does a DGX Station configured for an LLM cost?
The price depends on storage and service, not on the model: from €88,000 excl. VAT for the Supermicro with 6 TB of SSD to €110,500 excl. VAT for the ASUS with 8 TB and three years of on-site service (pi3g, 14 September 2026). The model you pick only changes which quantization format fits in memory.
Which LLM fits in a DGX Station?
With 748 GB of coherent memory, the DGX Station fits five of the seven open-weight models in the catalogue, from Qwen 3.8 27B in FP16 to Kimi K2.7 Code, 1,000 billion parameters, in NVFP4. The table at the bottom of the page gives the most precise format that fits for each.
Which runtimes are compatible with the DGX Station?
vLLM, llama.cpp or Triton Inference Server, depending on the framework. The model, format and GPU pairs measured by QDNA are published in the measurements section.
Which OEM versions of the DGX Station GB300 exist?
NVIDIA does not sell the DGX Station directly. It is built by Supermicro (Super AI Station ARS-511GD-NB-LCC), MSI (XpertStation WS300), Gigabyte (W775-V10-L01), HP (ZGX Fury G1n), ASUS (ExpertCenter Pro ET900N G3) and Dell (Pro Max with GB300), and configured by integrators such as Exxact and BIZON.
Why does the DGX Station price change from month to month?
Because of the memory shortage. At pi3g, the Supermicro Super AI Station went from €88,500 excl. VAT in July 2026 to €93,000 in August, then €88,000 in September, and the reseller expects about €125,000 excl. VAT in 2027.
Official sources
- DGX Station product page, NVIDIA
- DGX Station development guide, NVIDIA
- Prices and lead times by OEM, pi3g (page of 14 September 2026)
Which models fit on DGX Station?
The DGX Station offers 748 GB of coherent memory: five of the seven open-weight models in the catalogue fit, from 27 to 1,000 billion parameters. The table gives, for each, the most precise format that fits in memory.
| Model | Billion parameters | Most precise format that fits | Sizing page |
|---|---|---|---|
| GLM 5.3 | 744 | NVFP4 | GLM 5.3 on DGX Station |
| Kimi K2.7 Code | 1,000 | NVFP4 | Kimi K2.7 Code on DGX Station |
| Nemotron 3 Ultra | 550 | FP8 | Nemotron 3 Ultra on DGX Station |
| MiniMax M3 | 428 | FP8 | MiniMax M3 on DGX Station |
| Qwen 3.8 27B | 27 | FP16 | Qwen 3.8 27B on DGX Station |
See the full sizing matrix.