NVIDIA has announced a new 64GB configuration of its DGX Spark personal AI computer, giving developers, researchers and AI enthusiasts another option for running advanced AI workloads locally. NVIDIA announced the new configuration on October 2, 2026, with systems expected to become available from major hardware partners including Acer, ASUS, Dell, Gigabyte, HP and MSI starting October 23.
The new DGX Spark retains NVIDIA's GB10 Grace Blackwell Superchip, ConnectX-7 networking, DGX OS and the company's complete AI software stack. The major difference is the amount of unified memory: the new model comes with 64GB, compared with the 128GB configuration already available.
NVIDIA is positioning the machine as a compact desktop AI supercomputer capable of running AI models locally. According to the company, the 64GB version can run models with up to 100 billion parameters entirely on the device. This could be particularly useful for developers working with AI agents, local inference, fine-tuning, data science and applications where keeping data away from the cloud is important.
One of the most interesting features is the ability to connect two 64GB DGX Spark systems. Using NVIDIA's ConnectX-7 networking and Sync Cluster Assistant, the two machines can combine their memory into a 128GB pool, allowing larger workloads to run across the pair. NVIDIA says its internal testing showed up to 1.7× performance for two clustered 64GB systems compared with a single system in a Qwen 3.8 27B workload.
The new model starts at $4,999, making it NVIDIA's latest attempt to bring serious AI computing closer to individual developers rather than keeping it exclusively inside large data centers.
Comments