Discover NVIDIA DGX Spark Workstations and develop the next generation of AI models with on-premises AI

NVIDIA DGX Spark systems bring massive AI computing power right to your desk. Powered by the NVIDIA GB10 Grace Blackwell superchip, these compact systems deliver up to 1 petaFLOP of AI performance with extremely high energy efficiency. A full 128 gigabytes of coherent memory is shared by the CPU and GPU via a high-speed chip-to-chip connection. This allows you to run local AI models with up to 200 billion parameters—from DeepSeek to Llama to Gemma—on-premises at your own company in full compliance with data privacy regulations. This coherent memory architecture fundamentally distinguishes the NVIDIA DGX Spark systems from traditional systems with separate main memory and graphics memory. AI models do not need to be copied back and forth between main memory and VRAM, which results in enormous time and energy savings for large projects. We offer the NVIDIA DGX Spark platform through our partners, such as Gigabyte and ASUS.

Designed for the development of modern AI models and autonomous agents

With the pre-installed NVIDIA AI software stack and DGX OS, NVIDIA DGX Spark systems are ready to use right from the first boot. Local prototyping, fine-tuning, and inference run directly at your workstation, while you can later seamlessly deploy finished results to the data center or the cloud. As your project grows, you can link two NVIDIA DGX Spark systems via the integrated NVIDIA ConnectX-7 network card and work effortlessly with AI models featuring up to 405 billion parameters. The low power consumption compared to large multi-GPU systems makes this hardware the perfect choice for whisper-quiet, continuous operation in the office—without the need for complex cooling infrastructure.

Who the NVIDIA DGX Spark Systems Are Suited For

These systems are specifically designed for developers, data scientists, and research teams who want to get started right away with on-premises AI development without having to set up a fully customizable multi-GPU system from the start. They offer the ideal entry point for professional prototyping that can later be flexibly scaled up to larger infrastructures. Whether you choose the Gigabyte or ASUS model, the hardware always forms a tightly integrated, highly efficient unit. If you’re looking for even more performance for demanding AI projects, our AI PC buying guide will walk you through the process step by step to find the right setup.

Flexible Alternatives and the Next Level of Performance

If you need even more computing power and memory for massive enterprise models, the NVIDIA DGX Station is the next big step. As the ultimate desktop supercomputer featuring the NVIDIA GB300 Grace Blackwell Ultra superchip and 748 GB of coherent memory, it processes even the most complex workloads and autonomous agents right at your desk. If, on the other hand, you’re looking for a freely configurable system with modular, interchangeable graphics cards, you’ll find a wide selection of custom-built systems with up to 96 gigabytes of VRAM per card in our main category for AI workstations. Every system is thoroughly tested at our headquarters in Munich and prepared for B2B use. Business customers and public institutions also benefit from dedicated contact persons and special offers for competitive bids. Purchasing on account or flexible leasing models can also be easily arranged. Simply contact our B2B team directly by phone or email.