Nvidia adds 64GB DGX Spark at $4,999 while 128GB model rises to $6,950

Nvidia has spent the past year managing a memory shortage that reshaped its DGX Spark desktop AI supercomputer. The company launched the 128GB model at $3,999 in October 2025, then raised the price to $4,699 in February 2026, citing memory supply constraints. By October 2026, Nvidia's own store listed the 128GB Founders Edition at $6,950, and the system was out of stock when checked. Memory supplier Micron has warned that the crunch will worsen in 2027 and 2028, saying it has no line of sight to when supply and demand will return to balance.

NVIDIA announced on October 2, 2026 that a new 64GB unified-memory configuration of its DGX Spark will start at $4,999. The 64GB model will ship Friday, October 23, 2026, through six manufacturer partners: Acer, ASUS, Dell, Gigabyte, HP and MSI. Nvidia will not sell the 64GB version directly. It keeps the GB10 Grace Blackwell Superchip, DGX OS and the full Nvidia AI software stack, but halves the unified memory compared with the 128GB model.

PCMag reported on October 2, 2026 that the ongoing memory shortage prompted Nvidia to create the 64GB version, yet the new model costs $1,000 more than the original launch price for the 128GB models. The 128GB model now goes for $5,999, $6,499 and even $9,000 with 4TB storage, according to PCMag. HotHardware reported on October 2, 2026 that the 128GB version's price has increased significantly to $6,995, up from an MSRP of $4,699 and the original $3,999 launch price. Ground Truth reported on October 3, 2026 that at list prices the new 64GB model costs $1,951 less for half the memory, while two 64GB units cost $9,998 before cables and tax, more than one 128GB unit.

For developers who want to run large language models locally, the trade-off is both technical and financial. The 64GB DGX Spark supports up to 100-billion-parameter models fully on device, down from 200 billion on the 128GB version. Two 64GB units can be clustered with a QSFP cable, pooling memory to 128GB and expanding model support to up to 200 billion parameters, with twice the memory bandwidth and up to 1.7x the performance in Nvidia's Qwen 3.8 27B test. But clustering adds cost and complexity, and the memory ceiling remains the central constraint.

Key Facts

The 64GB DGX Spark starts at $4,999 and ships on Friday, October 23, 2026. It is sold by Acer, ASUS, Dell, Gigabyte, HP and MSI. The 64GB model retains the GB10 Grace Blackwell Superchip, DGX OS and the complete Nvidia AI software stack, along with an Nvidia ConnectX-7 networking chip and dual QSFP ports good for 200 Gbps of networking.

Memory capacity is the key difference. The 64GB configuration supports up to 100-billion-parameter models, while the 128GB model supports up to 200 billion. Two 64GB units can be connected with a QSFP cable through ConnectX-7, pooling memory to 128GB and supporting up to 200 billion parameters. Nvidia says the clustered pair delivers twice the memory bandwidth and up to 1.7x the performance of a single unit in its Qwen 3.8 27B test. A new Nvidia Sync Cluster Assistant detects connected units and auto-configures the ConnectX-7 network. An Nvidia Sync Model Launcher arriving at the end of October will download and launch Qwen3.8 27B and set up OpenCode.

Software support is broad. The system supports Ollama, vLLM, llama.cpp, LM Studio and PyTorch with CUDA out of the box. Nvidia highlights three workflows: running an AI agent around the clock, powering AI applications while the PC does other work, and scaling to two units. The most popular models, including Qwen 3.8 27B and Nemotron 3.5 Lightning 30B-A3B, still run on 64GB. Models around 80B parameters or larger, such as Qwen 3.8 Flash Next, are the big loss, according to HotHardware.

Nvidia's February price notice raised the Founders Edition from $3,999 to $4,699, citing memory supply constraints. The store later listed $6,950 with no explanation. PCMag reported that the 128GB model now goes for $5,999, $6,499 and even $9,000 with 4TB storage. HotHardware reported the 128GB version at $6,995, in line with partner systems such as the HP ZGX Nano G1n. Nvidia's own 128GB model remains out of stock, according to PCMag. On the AMD side, systems remain cheaper: the GMKtec AI Mini PC based on Ryzen AI Max+ 395 with 128GB is $3,650 and a 64GB version is $2,349, per HotHardware. AMD's alternative starts at $3,999, according to PCMag.

Analysis

What this really means is that Nvidia is trying to protect its DGX Spark business from a memory shortage it cannot control, but the fix creates a difficult value proposition. The 64GB model arrives at $4,999, which is $1,000 more than the 128GB model's original $3,999 launch price. For that higher price, buyers get half the unified memory and support for up to 100-billion-parameter models instead of 200 billion. The 128GB model, meanwhile, has climbed to $6,950 on Nvidia's store, roughly 75% above its launch price. The result is a cheaper option that costs more than the original flagship did.

The clustering story softens the blow but does not erase it. Two 64GB units can pool memory to 128GB and reach up to 200 billion parameters, with up to 1.7x performance in Nvidia's Qwen 3.8 27B test. Ground Truth reported that two 64GB units cost $9,998 before cables and tax, which is more than one 128GB unit at $6,950. It also requires a QSFP cable, two systems and the new Sync Cluster Assistant to auto-configure the ConnectX-7 network. For developers who already need 200-billion-parameter support, the 128GB model, when in stock, remains the simpler choice despite its higher price.

The bigger picture here is that memory supply is now a product design constraint, not just a cost line. Nvidia cut the DGX Spark's memory to 64GB to stem RAM price increases, according to HotHardware, yet the 128GB model's price still rose. Micron expects the crunch to worsen in 2027 and 2028, and says it does not have line of sight to when supply and demand will return to balance. That suggests Nvidia's configuration and pricing moves may not be temporary. The 64GB model could become the default entry point for local AI development on DGX Spark, while the 128GB model becomes a premium, supply-constrained option.

Competitive pressure makes the judgement sharper. AMD's alternative starts at $3,999, and HotHardware notes that AMD systems remain cheaper, such as the GMKtec AI Mini PC with Ryzen AI Max+ 395 at $3,650 for 128GB and $2,349 for 64GB. Nvidia's advantage is the Grace Blackwell chip, DGX OS, ConnectX-7 networking and the full software stack, including support for Ollama, vLLM, llama.cpp, LM Studio and PyTorch with CUDA. Nvidia is betting that its platform and partner channel justify the premium, even with less memory at the lower end.

Why It Matters

The 64GB DGX Spark matters because it changes who can buy into Nvidia's local AI platform. At $4,999 through Acer, ASUS, Dell, Gigabyte, HP and MSI, it is still a serious investment, but it is $1,951 less than the current $6,950 list price of the 128GB model. That gap could attract developers who want to run up to 100-billion-parameter models on device and who do not need 200-billion-parameter support right away. The system supports popular models such as Qwen 3.8 27B and Nemotron 3.5 Lightning 30B-A3B, along with Ollama, vLLM, llama.cpp, LM Studio and PyTorch with CUDA.

It also matters for the broader hardware market. The memory crunch is forcing Nvidia to segment its product line by capacity, and it is pushing prices up even as memory goes down. Micron's guidance that supply will stay tight into 2027 and 2028 means other vendors could face similar decisions. AMD's cheaper systems show that competition is ready to exploit any gap. The DGX Spark 64GB launch is therefore a test of whether Nvidia's software ecosystem and partner network can outweigh a straightforward price and memory comparison.

Next Up

Next up is the October 23, 2026 ship date, when Acer, ASUS, Dell, Gigabyte, HP and MSI are expected to release their 64GB DGX Spark systems. The Nvidia Sync Model Launcher is set to arrive at the end of October, bringing automated download and launch of Qwen3.8 27B plus OpenCode setup. Developers who want to cluster two units will also need the Sync Cluster Assistant, which detects connected systems and auto-configures the ConnectX-7 network.

Beyond October, the memory supply situation remains the wild card. Micron expects the crunch to worsen in 2027 and 2028, and has said it does not have line of sight to when supply and demand will return to balance. If that holds, the 64GB configuration may become the new baseline. The 128GB model, now at $6,950 and out of stock on Nvidia's store, could remain a premium option. AMD's cheaper systems, including the $3,650 GMKtec AI Mini PC with 128GB, will keep pressure on Nvidia to justify its pricing through software, networking and the Grace Blackwell platform.