Nvidia is moving a slice of AI development from remote data centers to the desk. Project DIGITS is a small desktop computer designed for people who want to experiment with AI models at home, including chatbots like ChatGPT and image generators.
The system starts at $3,000, debuted at CES 2025 in Las Vegas, and is scheduled to launch in May. It can work as a standalone PC or connect to a Windows or Mac machine.
A Desktop Built for Local AI Experiments
Project DIGITS is aimed at researchers, data scientists, and students. That audience matters because the device is not described as a general consumer gadget. Its purpose is to make serious AI development possible on personally owned hardware.
At CES on Monday, Nvidia CEO Jensen Huang described the system as “a cloud computing platform that sits on your desk.” The phrase captures Nvidia’s positioning: Project DIGITS is meant to give developers a local environment that resembles the kind of Nvidia-based infrastructure they may later use in cloud services or data centers.
That bridge is central to the product. Developers can build and test AI applications locally on Project DIGITS, then move those applications to larger systems that use similar Nvidia hardware. For people working with AI models, that could make early experimentation more direct before shifting work to larger deployment environments.
What Nvidia Put Inside Project DIGITS
The core of Project DIGITS is Nvidia’s new GB10 Grace Blackwell Superchip. The chip combines an Nvidia Blackwell GPU with a 20-core Grace CPU based on Arm architecture.
Nvidia developed the chip in partnership with MediaTek. Inside the Project DIGITS enclosure, it connects to 128GB of memory and up to 4TB of storage.
The memory figure is one of the most important technical details. AI model size is constrained by memory, and the source compares DIGITS’ 128GB of unified RAM with a high-power consumer GPU like the RTX 4090, which has only 24GB of VRAM.
That difference helps explain why Nvidia is presenting Project DIGITS as a local AI machine rather than just another desktop computer. More memory gives users room to run larger local AI models, and model size is a major practical limit for anyone trying to work outside remote data centers.
How Large a Model Can It Run?
A single Project DIGITS unit can reportedly run AI models with up to 200 billion parameters. Two linked units can handle models with 405 billion parameters.
Parameter count is a rough way to understand a model’s neural network size and complexity. Larger parameter counts generally require more memory and more computational power to run. Parameter size can also approximate capability, though models with different sizes can perform differently depending on how they were trained and architected.
The source points to several examples that help frame the range:
- Llama 3.1 70B, with 70 billion parameters, could probably run comfortably on Project DIGITS.
- Flux.1 dev, an AI image-synthesis model with 12 billion parameters, could also probably run comfortably.
- Llama 3.1 405B, with 405 billion parameters, may not run on a single unit.
Those examples show why the device is most interesting for local AI work that sits between ordinary consumer hardware and full remote infrastructure. It is not presented as a replacement for every large AI setup. It is presented as a compact system that can handle meaningful local experiments with sizable models.
The Software Stack Matters Too
Project DIGITS runs on Nvidia’s Linux-based DGX OS operating system. It also includes access to Nvidia’s AI software tools.
Those tools include the NeMo framework, which aids AI model development, and RAPIDS libraries, which developers use to create AI applications. The system can also run common AI development tools such as PyTorch, Python, and Jupyter notebooks.
That combination is important because local AI work is not only about hardware. Developers also need familiar tools for building, testing, and iterating. By supporting widely used development environments alongside Nvidia’s own software, Project DIGITS is designed to fit into existing AI workflows rather than forcing every task into a new toolchain.
Where It Fits Against Other Local AI Options
NVIDIA says the new computer starts at $3,000. The source notes that a fully maxed-out unit could cost considerably more.
There is also a comparison with Apple’s hardware. In a news release from October, Apple wrote that a Mac with an M4 Max chip can feature up to 128GB of unified memory and potentially run AI models up to 200 billion parameters in size, similar to the Project DIGITS computer.
At the time described in the source, a MacBook Pro with an M4 Max and 128GB of unified memory sells for about $4,699 in the US. That makes Project DIGITS potentially comparable in price and AI capability, although the true differences in AI performance would have to be revealed in lab testing.
The broader takeaway is straightforward: Nvidia is packaging local AI capability into a dedicated desktop system for people who build and test models. With Project DIGITS, the company is betting that more AI work will happen close to the developer before moving to cloud services or data centers.