Your own AI supercomputer.
In the EU. Never shared.
Whole dedicated NVIDIA DGX Sparks, billed by the minute. Your models, your data, your hardware, running in EU-Central, where no one trains on it but you.
EU-Central · live capacity shown in the console · $50 free on your first $100 top-up
Two ways to run a Spark
Rent by the minute in our racks, or test here, then take the hardware into your own building.
Rent in our cloud
Dedicated DGX Sparks in EU-Central, deployed in under a minute and billed only while they run.
- Per-minute billing: stopped instances cost nothing
- Root over SSH: a whole clean machine, bring your own stack
- API-first: deploy and manage from a single curl
- First top-up bonus: $50 free with your first $100 top-up
Own it in your office
The sane way to buy AI hardware is not to guess. Validate your workload on a rented Spark first, then bring the same machine on-premise.
- Test before you buy: your real workload, our racks
- EU-wide delivery: supplied and installed by us
- Managed service: monitoring, updates, support, training
- Fully yours: data never touches any cloud
The hardware
Every instance is one whole machine. Nothing is shared, virtualized, or oversubscribed.
- Superchip
- GB10 Grace Blackwell
- AI performance
- 1 PFLOP (sparse FP4)
- Memory
- 128 GB unified LPDDR5x
- CPU
- 20-core Arm
- Storage
- 1 TB NVMe
- Network
- ConnectX-7 · 200 Gb/s
- OS
- DGX OS 7 · Ubuntu 24.04
- Deploy time
- Under a minute
Deploy and connect
Every instance is a clean box with a hostname and key-based SSH. Root on real hardware, bring your own stack.
# deploy from the API curl -X POST https://api.gpuwerk.com/v1/instances \ -H "Authorization: Bearer $GPUWERK_TOKEN" \ -d '{"image":"dgx-os","ssh_public_key":"ssh-ed25519 AAAA..."}' # your shell on the box ssh -p 22001 root@ssh.gpuwerk.com # the whole GB10 is yours nvidia-smi && python -c "import torch; print(torch.cuda.get_device_name(0))"
Simple pricing
Per-minute billing against prepaid credit. No egress fees, no commitments. Your first $100 top-up comes with $50 free.
On-Demand
- Nodes
- 1× DGX Spark
- Memory
- 128 GB
- Storage
- 1 TB NVMe
- Setup fee
- $0
- Commitment
- None
On-Premise
- Nodes
- Yours, in your office
- Delivery
- EU-wide, installed
- Managed
- Updates · support · training
- Test first
- In our cloud, $50 free on first top-up
FAQ
The questions compliance teams and builders ask first.
Where is the hardware located?
Every DGX Spark in the fleet is physically located in EU-Central. Your data stays inside the EU, in transit, at rest, and while your models run.
Who can access my data?
You, and only you. Each instance is one whole physical machine with SSH-key-only access, no virtualization, no shared GPUs, and passwords are disabled fleet-wide. GPUwerk operates the infrastructure and never processes the content on your instance.
Is GPUwerk GDPR compliant?
We give you what you need to be compliant: hardware and operations in the EU, an Article 28 data processing agreement on request, no sub-processors for workloads, and no GPUwerk access to your workload content. GDPR compliance is a property of your processing and you remain the controller, so we supply the evidence rather than assert the conclusion on your behalf.
Can I buy DGX Sparks for my own office?
Yes. Test your workload on a rented Spark first, then take the hardware on-premise. We supply, install, and optionally manage DGX Sparks anywhere in the EU, support, updates, and training included. Ask for a quote.
What models can I run?
Anything that fits in 128 GB of unified memory: Llama, Qwen, DeepSeek and other open models, for inference or fine-tuning. We rent one node at a time, so 128 GB is the ceiling for a single job.
Start building on a Spark today.
$50 free with your first $100 top-up. Deployed in under a minute. No commitments, no egress fees.
Get started