Inside NVIDIA’s Compact AI Supercomputer: Project DIGITS

At CES 2025, NVIDIA introduced Project DIGITS, a groundbreaking AI supercomputer priced at $3,000 (£2,400). Compact yet incredibly powerful, it's designed for AI developers, researchers, and students seeking high-performance computing at the desktop level.

Traditionally, AI development relied on costly server clusters. Project DIGITS breaks this barrier, delivering a desktop-sized supercomputer built on a variant of the Grace Blackwell architecture—the same architecture that powers the world’s most advanced AI clusters.

A Mini Supercomputer with Big Performance

The heart of Project DIGITS is the GB10 Grace Blackwell Superchip, which merges a Blackwell GPU with a Grace CPU. This system-on-a-chip (SoC) offers 1 petaflop of AI performance at FP4 precision, perfect for model training, inference, and data science. Its 20-core Arm CPU and NVLink-C2C connectivity enable fast and efficient data transfer between components.

Unlike traditional AI servers that require elaborate cooling and infrastructure, Project DIGITS operates from a standard power outlet. It rivals high-end AI workstations, but its data-centre-grade internals in a desktop form factor make it unique and highly desirable for budget-conscious users.

Engineered for AI Developers

With 128GB of unified memory and up to 4TB of NVMe storage, Project DIGITS can run models with up to 200 billion parameters. Two units can be linked via NVIDIA ConnectX networking to support even larger models—up to 405 billion parameters.

Project DIGITS runs DGX OS, a Linux-based system optimized for AI workloads, offering seamless integration with the broader NVIDIA DGX ecosystem. This allows developers to train and fine-tune models locally before deploying them to larger infrastructure or the NVIDIA DGX Cloud.

Powerful AI Software Ecosystem

Users of Project DIGITS gain access to a rich AI software stack, including:

This ensures that developers can integrate Project DIGITS into existing workflows with minimal friction and maximum flexibility.

From Development to Deployment

Project DIGITS stands out with its scalable architecture. Developers can build, train, and test models locally and then deploy them across DGX Cloud or enterprise-level NVIDIA infrastructure. This seamless transition reduces deployment complexity and costs, especially for smaller teams and companies.

A New Era for Research and Education

Project DIGITS opens new doors for innovation across sectors:

  • Universities & Students – A budget-friendly, powerful AI research platform
  • Startups & Independent Researchers – A compact alternative to costly cloud compute
  • Enterprises – Ideal for fast prototyping and early-stage AI development

Jensen Huang, NVIDIA’s CEO, emphasized that putting supercomputing power on individual desks democratizes AI and empowers the next wave of innovation.

Fueling the Future of AI

Project DIGITS will accelerate development in several AI fields:

  • Natural Language Processing (NLP) – Build and train sophisticated chatbots locally
  • Computer Vision – Develop models for medical imaging, facial recognition, and more
  • Scientific Research – Analyze large datasets in genomics, climate modeling, and materials science
  • Robotics & Automation – Test and optimize robotics applications on-site

Availability and Pricing

Project DIGITS will be available starting May 2025 at a base price of $3,000 (£2,400). NVIDIA is collaborating with selected partners for global distribution. Interested developers can sign up for updates.

The system will be showcased at NVIDIA’s AI Developer Conference, where attendees can experience hands-on demos and deep-dive sessions.

Distilled

NVIDIA’s Project DIGITS marks a significant leap in AI computing accessibility. By merging affordability with performance, it empowers developers, researchers, and students to train and deploy large AI models without needing data centres. It’s a bold step toward redefining AI development and unlocking the next generation of innovation.