EU Flag Hosted in the EU • GDPR Compliant
NVIDIA DGX Spark

DGX Spark Cloud.

Rent high-performance GPU servers for training models, running agents, and handling complex engineering tasks.

Limited early-access slots available.

Built for AI Workloads

From training model weights to running large-scale applications.

Autonomous Agents

Run agent systems that require large VRAM and low-latency API responses.

AI Engineering

Get immediate SSH access. Rent a dedicated GPU instance and start running your code.

Model Training

Fine-tune or train large models. Optimized for multi-node setups and high-speed networking.

Hardware Specifications

Try NVIDIA Blackwell GPUs in the cloud before buying your own hardware.

ArchitectureNVIDIA Grace Blackwell
CPUNVIDIA GB10
VRAM128 GB Unified Memory
Storage4 TB NVMe
Networking400 Gbps InfiniBand
EnvironmentPre-installed CUDA, PyTorch, Docker

Pricing

Simple, transparent hourly pricing. No hidden fees.

Early Access

FREE
  • 128GB Unified VRAM
  • Root SSH access
  • Docker container
  • 1-hour trial limit

Regular

$0.8/hour
  • 128GB Unified VRAM
  • Root SSH access
  • Docker container
  • No time limits

FAQ

Have questions? Contact us at contact@sparkhosting.eu

DGX Spark is NVIDIA's latest GPU server built for heavy workloads like training AI models and running large applications.

It costs $0.80 per hour. If you join the waitlist, your first hour is free.

It is for developers, researchers, and small teams who need powerful GPU resources without long-term contracts.

Sign up for the waitlist. When we launch in August 2026, we will email you a link to book your server.

With 128GB of VRAM, you can run models up to 70 billion parameters without needing heavy quantization.

Instances are ephemeral. Data is wiped immediately after the instance is stopped. You must back up your data externally.

Our instances come with Docker, CUDA drivers, and popular frameworks like PyTorch pre-configured.