The archive · Product Ideas · Product decision · 2025
NVIDIA's Project Digits: a $3,000 desktop AI supercomputer that runs 200B-parameter models
NVIDIA unveiled a Mac-mini-sized computer with the GB10 Grace Blackwell Superchip, 128GB memory and 1 petaflop of AI performance, starting at $3,000.
NVIDIA
What it had to solve
Prototyping and fine-tuning large AI models has meant renting cloud GPUs or buying data-center hardware. NVIDIA wanted that work to happen on a desk: CEO Jensen Huang framed Project Digits as putting the Grace Blackwell platform 'on the desks of every data scientist, AI researcher and student.'
How it works
At CES 2025, NVIDIA announced Project Digits, a personal AI supercomputer priced from $3,000 and sized like a Mac Mini. Its heart is the GB10 Grace Blackwell Superchip, developed with MediaTek, which pairs an NVIDIA Blackwell GPU with a 20-core Grace CPU and 128GB of coherent unified memory, plus up to 4TB of NVMe storage.
NVIDIA said a single unit can run AI models of up to 200 billion parameters and delivers up to 1 petaflop of AI performance at FP4 precision. It runs Linux-based NVIDIA DGX OS with the full NVIDIA AI stack — development kits, orchestration tools and pre-trained models — and supports PyTorch, Python and Jupyter notebooks. Two units can be linked to handle models up to 405 billion parameters.
The pitch was a change of venue for AI work: instead of renting cloud capacity to prototype and fine-tune models, developers could do it locally, then deploy the same Grace Blackwell architecture to cloud or data-center infrastructure. Jensen Huang called it 'a cloud computing platform that sits on your desk,' aimed at AI researchers, data scientists and students.
Why it lands
- Compressing a data-center architecture to desk size let individual developers own the compute that previously required cloud rental.
- 128GB of unified memory was the enabling trick: enough to hold 200B-parameter models that a laptop cannot.
- Standard Linux, PyTorch and the full NVIDIA software stack meant no new skills were required to use it.
- The $3,000 price positioned it between a workstation and a server, making the 'personal supercomputer' category concrete.
What it did
The launch reframed the AI supercomputer as a $3,000 desk appliance: developers could prototype and fine-tune 200-billion-parameter models locally, then deploy to cloud or data-center infrastructure on the same Grace Blackwell architecture. Huang called it 'a cloud computing platform that sits on your desk.'
What you can take
When a technology is only reachable as a service, the product move is to make it ownable: compressing a data-center platform to desk size turned cloud-scale AI into a developer's personal machine.
Since then
Project Digits went on sale in May 2025 and shipped to developers that October as NVIDIA DGX Spark, with NVIDIA and partners delivering the desktop AI supercomputer; NVIDIA's product page continues to position it for local AI agents and large-model development.
Sources
- Nvidia announces $3,000 personal AI supercomputer called Digits
- Nvidia's Project Digits is a 'personal AI supercomputer'
spotted an error? The archive wants to know.
Your turn
You just read one. Describe the brief you are staring at, and see who has been given the same problem.
Free account · 3 free questions · no card