NVIDIA launches 64GB DGX Spark at $4,999 and expands 128GB version to $6,950, addressing AI workstation demand amid memory supply constraints.

Nvidia is introducing a more affordable variant of its DGX Spark, featuring half the memory and storage of the standard configuration. The GB10-based systems will be distributed exclusively through hardware partners including Acer, Asus, Dell, Gigabyte, HP, and MSI, with an expected retail price of approximately $4,999.

The new low-capacity SKUs follow Nvidia's price increase of the 128 GB DGX Spark to $6,950 on Friday—a nearly 75 percent jump from a year earlier. Rising memory costs are the primary driver of the increase. The 64 GB model positions itself significantly below Nvidia's top-specification DGX Spark and GB10 systems given the ongoing memory shortage, though it still commands 25 percent more than the 128 GB version retailed for at launch.

Pricing pressure on the GB10 platform has mounted steadily since the Project Digits concept debuted at CES last year. The appliance was originally projected to retail around $3,000 but eventually launched at $4,000. Even at that level, the DGX Spark delivered exceptional value—large quantities of GPU memory that would have cost tens of thousands of dollars just years earlier. At its debut, it held the distinction of being Nvidia's highest-capacity workstation GPU.

The 64 GB variant is less suited to memory-intensive AI workloads such as fine-tuning, but Nvidia argues it is adequate for local AI inference applications. The company points to models in the 26–35 billion parameter range—citing Qwen 3.8 27B as an example—as sufficiently capable for powering private agents.

Despite the reduced memory, the systems maintain unchanged memory bandwidth of 273 GB/s, achieved by using lower-capacity LPDDR5x memory modules rather than fewer modules overall. The 64 GB configuration retains the same 20-core Arm processor from MediaTek and onboard ConnectX-7 networking, enabling clustering of up to four GB10-based devices at 200 Gbps each.

Nvidia plans to release new software tools for DGX OS later this month to simplify cluster configuration and inference deployment—tasks previously requiring command-line tweaking and vLLM or TensorRT-LLM parameter adjustments.

The lower pricing of the 64 GB model does not bode well for the upcoming RTX Spark notebooks and mini PCs expected this fall. Built on the same GB10 silicon but targeting general-purpose computing—including AAA gaming, web browsing, and content creation—these devices were already anticipated to command premium prices, particularly 128 GB variants. With DGX Spark pricing now near $7,000, similarly configured RTX Spark notebooks could retail significantly higher, though Nvidia is expected to offer lower-memory versions for price-conscious buyers.

AMD's newly launched Gorgon Halo SoCs present competitive pressure, offering memory configurations from approximately 32 GB to 192 GB across the Ryzen AI Max+ 495 SKU lineup. Systems powered by these chips currently retail below Nvidia's 128 GB DGX Spark. However, testing has consistently demonstrated that the GB10 GPU delivers substantially higher performance for AI workloads.