Text & agentsMIT
DeepSeek V4 Flash DeepSeek · 304B A13B Intelligence index 50103GB · 87GB on DS4 · 131K ctx
Own your intelligence on your desk
Your private AI arrives configured
Frontier open models on hardware you own, with more under your desk than ever.
Access the whole lineup
128GBDGX Spark From €4,450 ex VAT NVIDIA
128GBAscent GX10 From €3,360 ex VAT ASUS
128GBPro Max with GB10 From €4,865 ex VAT Dell
128GBVeriton GN100 From €5,040 ex VAT Acer
128GBAI TOP ATOM From €4,620 ex VAT GIGABYTE
128GBZGX Nano From €4,586 ex VAT HP
128GBThinkStation PGX From €4,675 ex VAT Lenovo
128GBEdgeXpert MS-C931 From €4,620 ex VAT MSI
96GBTower 96 From €16,900 ex VAT Y
192GBTower 192 From €29,800 ex VAT Y*The eight desk boxes are the same NVIDIA GB10 Grace Blackwell platform: identical silicon, 128GB unified memory, identical throughput — they differ in chassis, storage, warranty channel and price. The two Y Towers are the tier above: RTX PRO 6000 builds, priced at order.
Read moreThe whole
stack, installedSee the stack
stack, installedSee the stack
- Local model server, quantised and tuned
- OpenClaw agent runtime, wired to your tools
- A ChatGPT-style app on your own network
128GB unified memory
Models that will not fit on a laptop
DeepSeek V4 Flash · 15.5–16.6 tok/s decode · 411 tok/s read
*Throughput verified hands-on on a single DGX Spark at UD-IQ3_XXS with 131,072 tokens of context. Before your machine ships we re-run the numbers on your exact unit and hand you the results. Zero tokens leave the building because inference never touches a network.
Read moreFrontier intelligence, answering to no one
There’s a model for everything
CodeOpenMDW
Laguna-S-2.1 Poolside · 118B A8B 71GB NVFP4 · 31 tok/s
Text & agentsOpen weights
MiniMax-M2.5 MiniMax · 230B A10B Intelligence index 44101GB · 26 tok/s
Text & agentsApache 2.0
Qwen3.5 122B Alibaba · A10B Intelligence index 3275.6GB NVFP4 · 262K ctx
Text & agentsOpen weights
Mistral Medium 3.5 Mistral AI Intelligence index 30~77GB at Q4
Text & agentsOpenMDW
Nemotron 3 Super NVIDIA · 120B Intelligence index 25~61GB at NVFP4
Text & agentsApache 2.0
gpt-oss-120b OpenAI Intelligence index 2459GB measured
Text & agentsOpen weights
Qwen3 Coder Next Alibaba · 80B Intelligence index 21~48GB at Q4
Text & agentsOpen weights
Devstral 2 Mistral AI Intelligence index 19~75GB at Q4
VideoFree < $10M ARR
LTX-2.3 Lightricks · 22B ~33GB working set
ImageOpen weights
FLUX.1-dev Black Forest Labs 95GB · fits, tight
ImageOpen weights
Qwen-Image-2512 Alibaba 63GB · measured
VoiceMIT
Chatterbox Resemble AI Clones from ~10s
MusicOpen weights
ACE-Step v1.5 ACE Studio Full tracks · ~10GB
Scores: Artificial Analysis Intelligence Index v4.1, nine independent evals. Memory figures are measured GGUF file sizes, not estimates. DeepSeek V4 Flash verified hands-on: 103GB resident, 18GB free, 131,072-token context on a single box.
Read moreWhat do you want to run on it?
All of it runs on the same machine. Pick one.
“Runs on a single DGX Spark.”
Poolside
Laguna-S-2.1 launch, on this exact hardware class
The honest version: if nothing runs while you sleep and nothing you type is sensitive, a subscription is the better deal and we will say so. The moment agents loop — or the data is not yours to share — the answer flips. Payback maths, worked through, at why-local/cost.
Token capacitymeasured every month
Your monthly capacity~180Mtokens, flat out, one box$1,940–$3,230/mo on the meter
agents by y.computer
Set it running.Come back to done.
Run 130+
open models
DeepSeekQwen3.5LagunaMiniMaxMistral
gpt-ossNemotronDevstralFLUX.1Qwen-Image
LTX videoWan 2.2ChatterboxKokoroACE-Step
WhisperParakeetSD 3.5Gemma 4Inkling
Agent runs shown are from the measured build: DeepSeek V4 Flash at 16 tok/s decode sustains roughly 4.2M tokens over an eight-hour overnight loop. Your workloads will differ — which is why every install ends with your numbers, not ours.
Read moreTen builds. Two tiers.
Eight OEMs ship the same NVIDIA GB10 — 128GB unified memory, identical measured throughput, pick on chassis and price. Above them, two Y Tower builds on RTX PRO 6000 GPUs for those who need more speed or native-FP8 fidelity.
The canonical build, NVIDIA Founders Edition 4TB storage NVIDIA direct warranty NVIDIA DGX Spark From €4,450 ex VAT The reference unit the whole platform is named for. If you want the machine every playbook, driver note and forum thread assumes, this is it.The value pick 1TB storage Lowest verified EU price ASUS Ascent GX10 From €3,360 ex VAT Identical silicon at the lowest price we have verified in the EU channel. The 1TB drive is the only trade — models stream from NAS happily.The enterprise channel 4TB storage Dell ProSupport Dell Pro Max with GB10 From €4,865 ex VAT For fleets and procurement departments: Dell's business channel, next-business-day support contracts and a chassis IT already knows how to file.Workstation pedigree 4TB storage HP ZGX toolchain HP ZGX Nano From €4,586 ex VAT HP's Z workstation lineage in GB10 form, with HP's global support network behind it. The conservative choice, in the best sense.The ISV-certified one 4TB storage ThinkStation reliability line Lenovo ThinkStation PGX From €4,675 ex VAT Lenovo's ThinkStation discipline applied to the same chip. Popular where certification lists and asset registers rule the purchase.The compact build 4TB storage Wide retail availability MSI EdgeXpert MS-C931 From €4,620 ex VAT MSI's take — easy to actually buy at retail in most regions, same 128GB pool, same throughput as every other box on this row.The US price leader 4TB storage Lowest verified US list GIGABYTE AI TOP ATOM From €4,620 ex VAT The cheapest verified US list price for the platform. GIGABYTE's AI TOP tooling comes along for the ride.The EU-stocked one 4TB storage Stocked in EU distribution Acer Veriton GN100 From €5,040 ex VAT Acer's build, reliably in stock with European distributors when others quote lead times. Same chip, same memory, same speed.The speed tier — RTX PRO 6000 96GB GDDR7 · 1.79TB/s Any wall socket · ~750W Y Tower 96 From €16,900 ex VAT The tier above the desk boxes: one RTX PRO 6000 Blackwell in a quiet tower. 6.6× a GB10's bandwidth — the 87GB flagship build, estimated at multiples of desk-box speed.The fidelity tier — two PRO 6000 192GB GDDR7 across two cards DeepSeek V4 Flash at native FP8 Y Tower 192 From €29,800 ex VAT The cleanest flagship on any home machine we know of: the official 167GB FP8 checkpoint fits whole. Priced at order, receipts on the invoice.
Unlock the free guide to running AI locally
Written by the people who install it — what fits in 128GB, where the payback turns, and the questions worth asking any vendor.