NVIDIA Unveils DGX Spark and DGX Station: Desktop Supercomputers Built for the AI Boom

Taipei, Taiwan — At Computex 2025, NVIDIA introduced two groundbreaking additions to its AI supercomputing arsenal: the DGX Spark and the newly redesigned DGX Station. These compact yet immensely powerful machines are engineered to meet the surging global demand for AI infrastructure—specifically in environments that require top-tier performance in a smaller, more accessible footprint.

The unveiling, led by CEO Jensen Huang, highlights NVIDIA’s strategic move to democratize supercomputing, bringing petaflop-class AI performance to researchers, developers, and creators beyond the confines of massive data centers.

“The world’s appetite for AI is exploding,” said Huang during the keynote. “With DGX Spark and DGX Station, we are putting supercomputing power directly into the hands of those building the future.”

DGX Spark: Petaflop Performance on Your Desktop

Designed in collaboration with NVIDIA’s global OEM partners, the DGX Spark is a marvel of miniaturization. Despite its desktop form factor, this compact system delivers over one petaflop of AI compute performance, thanks to its integration of NVIDIA’s cutting-edge Grace Blackwell architecture and H100-class GPUs.

Perfect for AI labs, R&D teams, and advanced engineering offices, the DGX Spark allows professionals to:

• Train and deploy large-scale AI models locally

• Run advanced simulations without relying on cloud resources

• Rapidly prototype and iterate across AI workflows with minimal latency

Its compact size and whisper-quiet thermal design make it ideal for on-prem environments where space and acoustics matter, without compromising performance.

DGX Station: A New Powerhouse for AI Professionals

Complementing the Spark is the revamped DGX Station, NVIDIA’s high-end workstation offering that blurs the line between desktop and supercomputer. Packed with multiple NVIDIA Grace Blackwell processors and a cluster of H100 Tensor Core GPUs, the DGX Station delivers the performance density of a rack-scale system—no server room required.

Built for:

• Advanced generative AI model training and fine-tuning

• Real-time 3D rendering and digital twin development

• AI-powered design, robotics, biotech, and industrial R&D

The DGX Station offers plug-and-play supercomputing for enterprises and researchers pushing the edge of what’s possible with AI.

Designed for the AI Era

Both systems are tightly integrated with NVIDIA’s full-stack AI platform, including:

• CUDA-X and TensorRT for optimized model development

• NVIDIA AI Enterprise for enterprise-grade reliability and manageability

• Omniverse and Modulus for digital twin and simulation workloads

This makes deployment seamless across enterprise and scientific applications, empowering teams to accelerate time-to-discovery and time-to-market for AI-driven solutions.

A Strategic Leap Forward

With DGX Spark and DGX Station, NVIDIA is reshaping the AI computing landscape. These systems bring unprecedented compute density and capability to the desktop, ensuring that labs, startups, and corporate innovation teams can compete at the forefront of AI development—without needing to scale up entire server clusters.

In the era of foundation models, generative AI, and real-time simulation, access to compute is power. NVIDIA’s latest hardware answers that need with elegance, efficiency, and a bold vision for AI at every desk.

TL;DR:

At Computex 2025, NVIDIA launched the DGX Spark, a desktop-sized AI powerhouse delivering over one petaflop of compute, and the revamped DGX Station, a workstation-class AI supercomputer featuring Grace Blackwell processors and H100 GPUs. These systems bring data-center-level AI performance into labs, studios, and enterprises—perfect for training, inference, and creative exploration in the age of generative AI.