NVIDIA Launches 64GB DGX Spark: Memory Halved, Starting at $4,999; 128GB Version Price Hiked to $6,950

As AI models become increasingly "lightweight," VRAM is becoming increasingly expensive. NVIDIA is addressing this contradiction with a product that "cuts memory."

On Friday, October 2 (US Eastern Time), NVIDIA announced the launch of the desktop AI computer DGX Spark with 64GB of unified memory, starting at $4,999. It will be brought to market on October 23 by OEM partners including Acer, ASUS, Dell, Gigabyte, HP, and MSI. The new machine retains the GB10 Grace Blackwell Superchip, DGX OS, the full NVIDIA AI software stack, and ConnectX-7 networking capabilities, but reduces unified memory from the existing 128GB to 64GB.

Interestingly, the starting price of this "scaled-down" DGX Spark is actually higher than the $3,999 launch price of the 128GB version last December; meanwhile, NVIDIA has further raised the price of the 128GB Founders Edition to $6,950. Media outlets pointed out that NVIDIA attributed the price hike to limited memory supply and rising costs.

This means that against the backdrop of AI server demand continuing to squeeze memory supply, "providing less memory" has become a way to lower the price threshold for local AI devices, but it does not mean the devices themselves have truly become cheaper.

Is 64GB Enough? NVIDIA Targets Local AI Agents

The core logic behind NVIDIA's adjustment is that an increasing number of open-source models can now run in smaller memory capacities.

NVIDIA stated that as open-source model capabilities improve and model sizes shrink, 64GB of unified memory is sufficient to support a range of local AI applications. The new DGX Spark can run models with up to 100 billion parameters locally on the device and supports workloads such as AI agents, inference, fine-tuning, data science, and edge development.

Tom's Hardware also noted that some of the latest high-performance dense models can already run in around 32GB of memory, although they remain limited in scenarios such as larger context windows. This means that the 128GB of memory originally prepared for "running large models locally" is not needed by all developers.

This is the practical basis for NVIDIA's launch of the 64GB version: for users primarily engaged in local inference and developing AI agents, rather than large-scale model training or fine-tuning, 64GB can cover a significant portion of workloads.

In its announcement, NVIDIA even summarized this trend as "the practicality of local AI is increasingly strengthening as technology continues to evolve," noting that the demand for running models locally is increasing as AI agents move from experimentation to daily development.

At the same time, local deployment offers attractions in terms of privacy and cost. Developers can process their own data and run AI agents directly on the DGX Spark without needing to call cloud-based models for every task.

Same GB10, 64GB Version Performance Does Not Shrink

From a hardware architecture perspective, the 64GB version is not an entirely new chip product.

NVIDIA stated that the new machine still uses the GB10 Grace Blackwell Superchip and retains the full DGX OS and AI software stack. Media outlets pointed out that the 64GB version also retains the original 20-core Arm CPU and 273GB/s shared memory bandwidth. Therefore, for models that can fit into 64GB of memory, the basic computing power of the two products has not changed due to the halving of memory capacity.

In other words, the focus of this change is not "cutting computing power," but rather cutting memory capacity that some users may not need.

The 64GB version still supports mainstream inference frameworks such as llama.cpp, Ollama, vLLM, and LM Studio, and comes pre-installed with the NVIDIA Agent Toolkit, CUDA-X AI libraries, and open models such as Nemotron.

NVIDIA's strategy is clear: if developers currently only need 64GB, they can purchase a lower-capacity machine first; if model scales continue to expand in the future, they can increase memory and computing power through clustering.

Two 64GB Units Can Be "Combined" into 128GB, Boosting Performance by Up to ~70%

Another key focus of this DGX Spark update is that NVIDIA has further strengthened multi-machine collaboration.

The 64GB version also includes a built-in ConnectX-7 network interface. Two devices can be directly connected via QSFP cables and, with the help of the NVIDIA Sync Cluster Assistant, automatically complete network configuration to form a local AI cluster. NVIDIA stated that two 64GB DGX Spark units can form a 128GB memory pool, supporting models with up to 200 billion parameters.

In NVIDIA's Qwen 3.8 27B test, after two 64GB machines formed a cluster, performance reached up to approximately 1.7 times that of a single machine.

NVIDIA will also launch the NVIDIA Sync Model Launcher in late October to further simplify the deployment process of models on single machines or clusters. For example, developers can use this tool to run Qwen 3.8 27B and integrate it with programming tools such as OpenCode.

This means that the product logic of DGX Spark is shifting from "a desktop AI supercomputer" to "a locally scalable AI node."

Of course, two machines do not equal one physically 128GB DGX Spark. Whether specific models can run across nodes still depends on software and workloads. Therefore, for tasks requiring large-memory single-machine operation, the 64GB version still has obvious limitations.

Amid VRAM Shortages, the 64GB Version at $4,999 Is Not Truly "Cheap"

What is truly worth paying attention to is actually the price.

NVIDIA's official starting price for the 64GB DGX Spark is $4,999, and this version will not have an NVIDIA-branded Founders Edition, but will be sold entirely through OEM partners. Partners include Acer, ASUS, Dell, Gigabyte, HP, and MSI, with specific configurations and prices potentially varying.

For comparison, when DGX Spark was initially launched in 2025, the official price for the 128GB version was $3,999; media outlets pointed out that NVIDIA subsequently raised the price of the 128GB Founders Edition to $4,699 in February 2026, and this latest adjustment brings it to $6,950.

Therefore, if only comparing historical launch prices, today's 64GB version is not only not cheaper than the 128GB version, but is actually $1,000 more expensive, representing a 25% increase.

However, compared to NVIDIA's current official price for the 128GB version, the 64GB version at $4,999 is still $1,951 cheaper than $6,950, but the cost is halving the memory capacity.

Citing relevant information, PC Watch reported that the price increase for the 128GB version is related to memory supply constraints and rising costs; Tom's Hardware pointed out that the actual selling price of 128GB GB10 systems in the current market has even reached approximately $7,000 to $9,000.

Therefore, another implication of this product adjustment is: AI computing power is sinking to local devices, but memory has become an important cost bottleneck in this trend.

Shifting from "Stacking Memory" to "On-Demand Expansion"

From a product strategy perspective, NVIDIA is not simply making DGX Spark a "low-spec version."

In the past, one of the core selling points of DGX Spark was its 128GB of unified memory, allowing developers to run large-parameter models on desktop devices; now, with advances in model compression and quantization technology, some models no longer require such large memory capacities.

Therefore, NVIDIA's solution has become: 64GB on a single machine satisfies mainstream local AI inference, and when memory demands increase further, expansion is achieved through multi-machine clustering.

The 64GB version will officially launch on October 23, with a starting price of $4,999. Meanwhile, the price hike of the 128GB version to $6,950 makes this new product particularly notable—it is both NVIDIA's attempt to lower the hardware threshold for local AI, and a mirror reflecting the current AI industry's contradiction of "strong demand for computing power and tightening memory supply."