Mini desktop supercomputer coming this year — powerful enough to run advanced AI models and small enough to fit in your bag

Nvidia has unveiled a line of artificial intelligence (AI) supercomputers that can deliver unprecedented processing power in a portable, desktop-friendly chassis.

Previously dubbed Project Digits, the powerful machines first revealed at CES 2025 in January have been rebranded as the DGX Spark and DGX Station machines. These computers are powered by Nvidia’s Blackwell Ultra platform and promise up to a petaFLOP in processing power — upwards of 1,000 times faster than the best laptops or high-end desktop PCs.

Blackwell Ultra is designed for massive-scale AI training and testing, and the DGX machines promise to put that power into the hands of data scientists, AI researchers and students at a relatively affordable price point. It’s the equivalent of putting the power of a data center into a computer small enough to fit on your desk.

The DGX Spark is a little box smaller than a laptop that you could easily tuck away on a corner of your desk or fit into your bag. It stands just under 2 inches (5 centimeters) in height and slightly under 6 inches (15 cm) in width, and at its core is the GB10 Grace Blackwell Superchip, capable of delivering 1,000 AI TOPS (trillions of operations per second).

Related: The 9 most powerful supercomputers in the world right now

It also comes with 128 gigabytes of unified system memory and Nvidia’s full stack of AI software, including a number of tools, libraries as well as some pretrained models. This includes things like the CUDA Deep Neural Network (cuDNN) library for enhancing neural network layers during training and inference and a pre-trained SegFormer model. The Nvidia version of the DGX Spark is available to reserve online starting at $3,999, although the company has said other models will soon be available from manufacturers like ASUS, Dell and Lenovo.

Supercomputing power in a desktop tower chassis

The DGX Station is the Spark’s larger, more powerful sibling and is closer in size to a professional workstation.

Built around the GB300 Grace Blackwell Ultra Desktop Superchip, it contains a staggering 748 GB of "large coherent" memory — memory which can be accessed by more than one processor at a time.

It also features Nvidia’s ConnectX-8 SuperNIC, which enables network connectivity at a blistering rate of up to 800 gigabits a second — fast enough to download approximately five 4K movies in a second.

It also uses the NVLink-C2C Interconnect to connect internal components at 900 gigabytes per second. The DGX Station is a powerhouse designed to execute large-scale AI training and inferencing workloads from the comfort of your desktop, without having to access additional resources through the cloud.

The DGX Station isn’t currently available to reserve, but Nvidia has indicated it will be available later in 2025 from partners like Asus, Dell, HP, Lambda and Supermicro.

Jensen Huang, Nvidia’s founder and CEO, said the new DGX machines represent a natural next step in AI development.

"AI has transformed every layer of the computing stack. It stands to reason a new class of computers would emerge, designed for AI-native developers and to run AI-native applications,” he said in a statement. "With these new DGX personal AI computers, AI can span from cloud services to desktop and edge applications."