Nvidia to launch personal AI supercomputer at $3,000

Nvidia plans to bring the AI computing capabilities of its powerful Blackwell chips to PCs, enabling AI researchers, data scientists and students to develop and run inference on large AI models using their own desktop system. 

The Taiwan-based chipmaker announced Monday at the Consumer Electronics Show (CES) 2025 that it will release a personal supercomputer named Project Digits. It will be powered by the GB10 chip, which leverages the same Grace Blackwell architecture as the more advanced B200 chips used in AI supercomputers and data centres. 

Nvidia said Project Digits can deliver one petaflop of AI computing performance, allowing it to handle AI models with up to 200 billion parameters. Researchers can also link two Project Digits computers using NVIDIA ConnectX to run models with up to 405 billion parameters.  

One petaflop is equal to 1,000 teraflops. This means Nvidia’s AI PC will be significantly faster than any existing desktop computer or workstation, which typically deliver a few teraflops (trillion floating-point operations per second) of processing power at most. 

Project Digits will be available from May at a starting price of $3,000.  

Use of AI chips in desktop and laptop PCs is not new. For instance, Intel’s Neural Compute Stick (NCS) has been used by researchers with regular PCs to build and train AI models for several years. Also, many new laptops now have integrated neural processing units (NPUs) for AI tasks. However, both the NCS and these integrated NPUs are typically designed for less demanding AI applications. 

AI chips like Blackwell and Hopper have specialized hardware and optimized architectures that allow them to handle complex and resource intensive AI models more efficiently than traditional chips. These chips have specialized hardware accelerators to efficiently perform specific AI operations, such as matrix multiplication and convolution. They also have dedicated memory and low-precision arithmetic units, enabling fast data access and improved energy efficiency. 

“AI will be mainstream in every application for every industry. With Project Digits, the Grace Blackwell Superchip comes to millions of developers. Placing an AI supercomputer on the desks of every data scientist, AI researcher and student empowers them to engage and shape the age of AI,” said Jensen Huang, founder and CEO of Nvidia. 

Project Digits will run Linux-based Nvidia DGX OS and will provide researchers access to Nvidia’s extensive library of development kits, orchestration tools, and frameworks to test and build AI models that can be deployed on the Nvidia DGX Cloud. 

Traditionally known for its gaming GPUs, Nvidia has seen unprecedented growth in recent years fueled by rising demand for its AI chips, particularly Hopper H100, which is widely used for training and running models such as GPT-4. 

In March 2024, Nvidia unveiled the more advanced Blackwell B200 chips. However, its production was delayed by a few months due to a design flaw resulting in overheating issues. Shipments reportedly started in December 2024 after the issue was resolved. 

Last year, Nvidia surpassed Apple and Microsoft to briefly become the world’s most valuable company.  The chipmaker is currently the second most valuable company with a market capitalization of $3.66 trillion. 

Image credit: Nvidia