Super AI Station ARS-511GD-NB-LCC
The only liquid-cooled GB300 Station — near-silent, built for 24/7 duty, with an optional 5U rackmount conversion.
Price
$125,990
Newegg US · liquid-cooled variant · UK/EU quote-only
verified 2026-07 · supply & lease options in every proposal
Supermicro- 748 GB
- Coherent memory
- 7.1 TB/s
- HBM3e bandwidth
- 20 PFLOPS
- Sparse FP4 compute
- 1,600 W
- Max power
252 GB HBM3e + 496 GB LPDDR5X
26× a DGX Spark
NVFP4, Blackwell Ultra
standard UK 13 A socket
Supermicro's build is the engineering outlier: a serviceable closed-loop liquid cooler (cold plates and radiator) makes it the quietest Station and the one pitched for continuous production duty — and it is the only one that converts to a 5U rack unit with an optional kit, so the deskside pilot can move to the comms room without a repurchase.
US pricing is listed at $125,990 (Newegg). An EU reseller figure of ~€82,444 was observed but could not be reconciled (likely ex-VAT, pre-config) — treat European pricing as quote-only.
vs its siblings: Only liquid-cooled Station (near-silent, 24/7-rated); tower-to-5U-rack conversion kit; Supermicro also offers a GB200 HPC sibling.
Memory, to scale
748 GB model-visible · bandwidth is the speed limit
GPU HBM3e
252 GB · 7.1 TB/s
HBM3e
CPU LPDDR5X
496 GB · 396 GB/s
LPDDR5x
For scale
DGX Spark — 128 GB @ 273 GB/s
RTX PRO 6000 — 96 GB @ 1.79 TB/s
Mac Studio M3 Ultra — 512 GB @ 819 GB/s
DGX Station GB300 — 748 GB coherent
Stations aggregate over dual 400 Gb ConnectX-8 Ethernet (RoCE) — no inter-station NVLink.
What it actually runs
Declared from research and benchmarks, not computed marketing — tokens-per-second figures are cited where a real measurement exists.
- Kimi K2.6 (1T)INT4 nativefits594 GB — the whole trillion-param model, one box
- GLM-5.24-bitwith headroom~390 GB with room for 1M-token context work
- Qwen3.5 397BFP8 officialwith headroom397 GB — full-precision-class quality
- MiniMax M3FP8with headroom428 GB — multimodal at speed
- Mistral Large 3FP8fits675 GB — fits with modest KV headroom
- DeepSeek V4 Pro (1.6T)INT42+ units linked~800 GB — two stations over 400 GbE
The full sheet
Compute
- Superchip
- NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
- GPU
- Blackwell Ultra — up to 20 PFLOPS sparse FP4 (NVFP4)
- CPU
- NVIDIA Grace — 72× Arm Neoverse V2
- Interconnect
- NVLink-C2C 900 GB/s (CPU ↔ GPU, coherent)
- Multi-user
- MIG — up to 7 isolated GPU instances
Memory
- GPU memory
- 252 GB HBM3e @ 7.1 TB/s (7 of 8 stacks enabled)
- CPU memory
- 496 GB LPDDR5X (4× 128 GB SOCAMM) @ 396 GB/s
- Total coherent
- 748 GB (originally announced as 784 GB — shipping silicon is 748 GB)
- Model capacity
- Up to ~1T params FP4 — Kimi K2.6 native INT4 (594 GB) fits whole
Storage
- NVMe
- 4× M.2 NVMe (2× 960 GB system + 2× 1.92 TB data typical)
Networking & expansion
- SuperNIC
- ConnectX-8 — 2× QSFP112 @ 400 Gb/s (800 Gb/s aggregate)
- Ethernet
- 1× 10 GbE + 1× 1 GbE BMC
- Expansion
- 3× PCIe Gen5 slots · optional RTX PRO Blackwell display GPU
- Clustering
- Multi-node over dual 400 GbE (RoCE) — no inter-station NVLink
Software
- OS
- Ubuntu / DGX OS base + NVIDIA AI Enterprise ecosystem, NIM microservices
- Windows
- 'DGX Station for Windows' announced — Q4 2026
Where it shines
- Near-trillion-parameter open weights on one desk, one wall socket
- 252 GB of HBM3e at 7.1 TB/s — data-centre bandwidth, deskside
- MIG partitions one station into up to 7 governed instances
- Liquid cooling: quietest of the family, rated for continuous duty
- 5U rack conversion — pilot on a desk, promote to the comms room
The trade-offs
- Capital cost of a small fleet of cars — the audit must justify it
- 1,600 W under load: power, heat and placement need planning
- Anything spilling past the 252 GB HBM tier runs at LPDDR speed (396 GB/s)
- Arm platform — x86 containers need rebuilding
Buy this box for
Understanding NVIDIA DGX Station
GB300 Grace Blackwell Ultra Desktop Superchip
The DGX Station is a data-centre superchip in a tower: NVIDIA's GB300 pairs a Blackwell Ultra GPU (252 GB HBM3e at 7.1 TB/s) with a 72-core Grace CPU (496 GB LPDDR5X at 396 GB/s) — 748 GB of coherent memory connected by 900 GB/s NVLink-C2C. That is enough to serve near-trillion-parameter open weights at FP4 — Kimi K2.6's native INT4 checkpoint is 594 GB — from one 1,600 W wall socket. MIG partitioning splits it into up to 7 isolated instances for a team.
Like the Spark, it is a reference platform: ASUS, Dell, HP, MSI, Gigabyte and Supermicro each build their own version, with integrator builds from Exxact — same silicon, different cooling, storage and support. (BOXX and Lambda were named at launch but have no shipping SKU as of July 2026.) Published pricing runs $85,000 (MSI's MSRP) to ~$126,000; several OEMs are quote-only, and the one verified UK sticker is ASUS's £117,599.99 inc VAT. Lead times run 4–12 weeks.
Worth knowing: the shipping silicon has 252 GB of HBM3e (7 of 8 stacks enabled) and 748 GB total — less than the 288 GB/784 GB originally announced and still widely misquoted. Stations cluster over dual 400 Gb ConnectX-8 Ethernet rather than NVLink, and a Windows edition ('DGX Station for Windows') arrives Q4 2026.
Siblings on the same silicon
MSI
XpertStation WS300
The value flagship — the lowest published MSRP in the GB300 family, aimed squarely at sovereign-AI and private-lab buyers.
- Memory
- 748 GB
- Bandwidth
- 7.1 TB/s
- AI perf
- 20 PFLOPS FP4
$85,000 MSRP
Street $96,996 (CDW) · UK quote-only — lowest published price in the family
ASUS
ExpertCenter Pro ET900N G3
The only GB300 Station with a verified UK sticker price — £117,599.99 — plus PCIe 6.0 data slots and a public Windows commitment.
- Memory
- 748 GB
- Bandwidth
- 7.1 TB/s
- AI perf
- 20 PFLOPS FP4
£117,599.99 inc VAT
The only verified UK list price in the family · US $99,999
Gigabyte
W775-V10-L01
The straight-down-the-line reference tower — listed US pricing at $123,500 through server integrators.
- Memory
- 748 GB
- Bandwidth
- 7.1 TB/s
- AI perf
- 20 PFLOPS FP4
$123,500
US integrator list (Rackmount Pro) · UK quote-only
Dell
Pro Max with GB300
The enterprise-services build — 16 TB of storage and a bundled RTX PRO 2000 display GPU, priced only by conversation.
- Memory
- 748 GB
- Bandwidth
- 7.1 TB/s
- AI perf
- 20 PFLOPS FP4
Quote-only
Dell declined to publish pricing — family range $85k–$126k is the anchor
Exxact
Valence VWS-158270643
The build-to-order Station — a US integrator's take on the GB300 platform, configured to spec rather than sold off a shelf.
- Memory
- 748 GB
- Bandwidth
- 7.1 TB/s
- AI perf
- 20 PFLOPS FP4
Quote-only
Build-to-order US integrator — price against the $85k–$126k family range
HP
ZGX Fury AI Station G1n
The Windows-first Station — 'the most powerful Windows AI PC ever built', arriving late 2026. Register interest, don't wait on it.
- Memory
- 748 GB
- Bandwidth
- 7.1 TB/s
- AI perf
- 20 PFLOPS FP4
TBA — pre-order
Hardware ~Aug 2026, Windows edition Q4 2026 · press expectation ~$94k+
Sources & verification
Specifications and prices verified 2026-07 against the sources below. The memory shortage is repricing this market monthly — we re-verify at quote.
Compare against the rest of the catalogue, or have us spec it against your workloads.
Back to catalogueSpec the Super AI Station ARS-511GD-NB-LCC for your workload
We'll size it against your real token volumes, quote supply and lease options, and stand up the models with evals that prove they do the job — or tell you honestly that a different box (or the API) wins.