NVIDIA official newsroom (nvidianews.nvidia.com/bios/jensen-huang)
Nvidia launches DGX Spark 64GB configuration with hardware partners
The cheaper desk-side AI box arrives as memory shortages push the 128GB model to $6,950
Nvidia has a new, smaller entry point for its personal AI supercomputer. The company debuted a 64GB configuration of DGX Spark on October 2, 2026, priced at approximately $4,999.
The timing matters. Memory prices are soaring, supplies are short, and the original 128GB model now retails for $6,950 after a price increase of nearly 75%.
The gap between the two models is now close to $2,000.
What you get for the money
The engine is the GB10 Grace Blackwell Superchip. It pairs a 20-core Arm CPU with a Blackwell GPU on a single package.
Memory bandwidth stays at 273 GB/s.
Networking also carries over. The system includes ConnectX-7, Nvidia’s high-speed networking hardware, which lets users link up to four units together in a cluster.
With half the memory, the 64GB version is aimed at mid-sized model inference. Nvidia’s target range for this box is models with roughly 26 to 35 billion parameters.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Partners and software
The 64GB configuration is launching with established hardware makers including Acer, ASUS, Dell, GIGABYTE, HP, and MSI.
Lenovo appears in the broader DGX Spark ecosystem. It was not explicitly named as a partner for this particular SKU.
Nvidia plans to release updated DGX OS software later in October 2026. That update is designed to make cluster setup easier and to simplify deploying inference models.
What this means for developers, enterprises, and the market
The most direct beneficiaries are developers and companies running local inference. These are teams that want models on their own hardware rather than renting cloud capacity.
The model-size ceiling is the trade-off to watch. A box tuned for 26 to 35 billion parameters works well for many mid-sized models, but larger ones will push buyers toward the 128GB version or a cluster.
The October DGX OS update is arguably as important as the hardware. Clustering is only attractive if it is easy, and Nvidia is explicitly targeting that friction.
The risk sits with memory itself. If prices keep climbing, even the 64GB configuration’s pricing could come under pressure, and the gap between the two models might shift again.