NVIDIA has unveiled the DGX Spark and DGX Station at CES, bringing unprecedented AI supercomputing power to the desktop. These deskside systems are powered by the NVIDIA Grace Blackwell architecture, offering petaflop-level performance and large unified memory. The DGX Spark can handle models up to 100 billion parameters, while the DGX Station tackles massive 1-trillion-parameter models.

This launch is a pivotal moment for AI development, allowing researchers and developers to work with frontier open-source models locally without relying solely on cloud infrastructure. The ability to prototype and iterate on such powerful hardware locally significantly accelerates the development cycle for complex AI applications. Furthermore, the seamless scalability from these deskside units to the cloud offers a flexible hybrid infrastructure, democratizing access to high-end AI compute resources and empowering innovation across various industries.