Eight boxes,one chip
Every machine here is the same NVIDIA GB10 Grace Blackwell platform: identical compute, identical 128GB unified memory, identical throughput. You are choosing a chassis, a storage size, and a warranty channel — not a performance tier.
- Superchip
- NVIDIA GB10 Grace Blackwell
- Unified memory
- 128GB LPDDR5x, shared CPU/GPU
- Usable for models
- ~119–121GB after OS reserve
- CPU
- 20-core Arm (10 Cortex-X925 + 10 A725)
- AI compute
- Up to 1 PFLOP sparse FP4
- Clustering
- 2 nodes over ConnectX-7 (256GB pool)
- Runs
- DeepSeek V4 Flash 3-bit, MiniMax-M2.5, Qwen3.5 122B, gpt-oss-120b
DGX SparkFrom $4,699 ex sales taxNVIDIA · 4TBSee this system
Ascent GX10From $3,999 ex sales taxASUS · 1TB / 4TBSee this system
Pro Max with GB10From $6,331 ex sales taxDell · 4TBSee this system
Veriton GN100From $5,521 ex sales taxAcer · 4TBSee this system
AI TOP ATOMFrom $4,400 ex sales taxGIGABYTE · 4TBSee this system
ZGX NanoFrom $6,353 ex sales taxHP · 4TBSee this system
ThinkStation PGXFrom $5,849 ex sales taxLenovo · 4TBSee this system
EdgeXpert MS-C931From $4,815 ex sales taxMSI · 4TBSee this systemWhichever box, the stack is ours
Models quantised and installed, the agent runtime, your private chat app, everything logged in and working — flat €2,000 on top of the hardware, which we pass through at cost.