NVIDIA is adding a 64GB configuration to its DGX Spark local-AI platform, with systems coming this month from Acer, ASUS, Dell, Gigabyte, HP and MSI. The new configuration keeps the same GB10 Grace Blackwell Superchip, DGX OS and NVIDIA AI software stack used by the 128GB model while offering developers another memory option for running AI workloads locally. NVIDIA announced the change on October 2, describing the system as a way to run increasingly capable AI agents and open models directly on a desktop machine.
The 64GB systems are being offered exclusively through NVIDIA’s manufacturing partners rather than as a new NVIDIA-branded configuration. NVIDIA says the hardware can run models with up to 100 billion parameters entirely on the device, depending on the model and workload, while retaining the same core platform as the 128GB version. The company is positioning the smaller-memory configuration for developers, researchers and AI enthusiasts who want local inference and agentic applications without requiring the full memory capacity of the existing system.
The new option also retains the networking hardware needed to connect multiple DGX Spark systems. NVIDIA says users can link systems together with its Sync Cluster Assistant, allowing workloads that exceed the capacity of a single machine to be distributed across multiple nodes. NVIDIA is also introducing Sync Model Launcher, which can automate downloading and launching supported models across one or more systems, reducing some of the setup work involved in building a local AI cluster.
That clustering capability gives the 64GB configuration a different role from simply being a cheaper version of the existing machine. A single system can handle models that fit within its available unified memory, while multiple systems can be combined when a workload requires additional memory and compute resources. NVIDIA says the 64GB configuration can therefore serve as an entry point for developers who may later expand to a multi-system setup rather than committing to the maximum memory configuration from the outset.
The underlying hardware remains centered on NVIDIA’s GB10 Grace Blackwell Superchip, with the 64GB configuration using the same DGX OS and NVIDIA AI software environment as the 128GB system. That software stack is an important part of the platform’s positioning because DGX Spark is intended specifically for local AI development rather than functioning as a conventional desktop PC. The company has increasingly focused on running models and AI agents locally, where developers can work without sending every inference workload to a cloud service.
Pricing and availability information reported alongside the announcement points to a starting price of $4,999 for the 64GB systems, with launch expected on October 23 through the named hardware partners. NVIDIA’s announcement itself confirms availability during October but does not provide the full partner-by-partner pricing breakdown, so the exact retail configurations will depend on the individual manufacturers. The 128GB DGX Spark configuration remains available for workloads that require more memory, including larger models and memory-intensive development tasks.
The timing also reflects a broader push toward running AI workloads on dedicated local hardware. NVIDIA says open models are becoming capable enough to fit on increasingly compact systems, while developers are using local machines for agent development and other workloads that previously relied more heavily on cloud infrastructure. The 64GB DGX Spark gives that strategy another hardware tier without changing the underlying Grace Blackwell platform or its software environment.
For developers considering the platform, the main tradeoff is memory capacity rather than a different generation of processor or software. The 64GB model provides less room for the largest local models than the 128GB version, but it retains the platform’s core architecture and can be connected to additional Spark systems when more capacity is needed. That makes the new configuration an expansion of NVIDIA’s local-AI hardware lineup rather than a replacement for the existing DGX Spark.
