Nvidia has launched the DGX Spark 64GB, a desktop AI computer priced at $4,999, and raised its 128GB sibling to $6,950. Both prices were set on 2 October 2026, with retail sales starting 23 October through six hardware partners. The 128GB model launched at about $4,000.
What the DGX Spark 64GB Costs and When It Ships
The 64GB model carries a manufacturer’s price of $4,999 and reaches buyers on 23 October 2026. Nvidia, the Santa Clara chip designer behind the Grace Blackwell platform, says the configuration is available exclusively through participating OEM partners rather than directly from Nvidia.
The partners named for the launch are Acer, ASUS, Dell, Gigabyte, HP and MSI. Nvidia’s own product page still lists the 64GB configuration as coming soon with a notification sign-up, so the 23 October date comes from the launch coverage rather than from a date published by Nvidia.
What the 64GB Model Gives Up, and What It Keeps
The cut is to memory and storage, not to the processor or the interconnect. Every figure below for the chip, bandwidth and networking is identical across the two configurations, which is the detail that decides whether the cheaper box is worth it.
| Specification | DGX Spark 64GB | DGX Spark 128GB |
|---|---|---|
| Price | $4,999 | $6,950 |
| Unified memory | 64GB LPDDR5X | 128GB LPDDR5X |
| Memory bandwidth | 273 GB/s | 273 GB/s |
| Largest model, per Nvidia | Up to 100 billion parameters | Up to 200 billion parameters |
| Processor | GB10 Grace Blackwell Superchip | GB10 Grace Blackwell Superchip |
| Stated AI performance | Up to 1 petaFLOP at FP4 | Up to 1 petaFLOP at FP4 |
| Networking | ConnectX-7 at 200 Gbps | ConnectX-7 at 200 Gbps |
Both configurations use the same GB10 Grace Blackwell Superchip, which Nvidia’s specification page lists as a 20-core Arm processor — ten Cortex-X925 and ten Cortex-A725 cores — paired with a Blackwell GPU, in a 1.2kg chassis rated at 140W.
Storage is the less-publicised reduction. The Register reported that the 64GB model halves storage as well as memory; Nvidia’s page lists up to 4TB of self-encrypting NVMe M.2 storage for the DGX Spark line without separating the configurations.
Why Two 64GB Boxes Cost More Than One 128GB
Clustering does not rescue the cheaper model on price. Nvidia supports linking up to four DGX Spark systems over ConnectX-7, for workloads of up to 700 billion parameters, which makes buying two 64GB units a plausible route to 128GB of memory.
The arithmetic argues against it. Two 64GB units at $4,999 each come to $9,998, which is $3,048 more than the $6,950 Nvidia now charges for a single 128GB machine holding the same total memory. The pair does add a second processor and a second petaFLOP of rated performance, but a model that needs more than 64GB of memory does not run faster for being split across two boxes connected at 200 Gbps.
At $4,999 the 64GB configuration also costs roughly 25% more than the 128GB machine did when the line launched at about $4,000.
How Large a Model It Actually Runs
Nvidia’s headline figure and its practical guidance do not match, and the difference is large enough to matter before buying. The specification page puts the 64GB configuration at up to 100 billion parameters.
The working range Nvidia describes is far lower. The Register reported that Nvidia positions the 64GB model for “models in the 26-35 billion parameter range, Qwen 3.8 27B for example”, aimed at local inference for private agents.
The gap is a question of precision rather than contradiction: a 100-billion-parameter model fits in 64GB only when heavily quantised, while a 27-billion-parameter model fits with room for context and other work. Buyers sizing a purchase against a specific model should work from the lower range.
Why the 128GB Price Rose 75% Since Launch
As of 4 October 2026, the 128GB DGX Spark sells for $6,950, nearly 75% above the roughly $4,000 it launched at. Nvidia attributes the increase to limited memory supply.
The pressure is industry-wide rather than specific to this product. DRAM manufacturers have shifted wafer capacity toward high-bandwidth memory for AI data centres, where margins are higher, which has tightened supply and raised prices for the conventional memory that desktops and laptops use. Our report on CXMT’s G5 DRAM and its missing yield figures covers one manufacturer’s response to that demand.
Nvidia has not said whether the 64GB price will hold, and has given no date for any review of either price. The company’s other recent product news is covered in our report on Nvidia OpenShell and the Sentry hardware requirement.
Frequently Asked Questions
How Much Does the DGX Spark 64GB Cost?
Nvidia set the price at $4,999 on 2 October 2026. That is about 25% more than the 128GB configuration cost when the line launched at roughly $4,000.
When Can You Buy the DGX Spark 64GB?
Launch coverage puts retail availability at 23 October 2026 through Acer, ASUS, Dell, Gigabyte, HP and MSI. Nvidia’s own page still lists the configuration as coming soon.
Is the 64GB Model Slower Than the 128GB?
Not on the published figures. Both list the same GB10 Grace Blackwell Superchip, the same 273 GB/s memory bandwidth and the same rated 1 petaFLOP at FP4 precision.
Should You Buy Two 64GB Units Instead of One 128GB?
Two 64GB units cost $9,998, which is $3,048 more than a single 128GB unit with the same total memory. The pair adds a second processor, but splitting one large model across two linked boxes does not make it run faster.
Why Did the 128GB DGX Spark Get More Expensive?
Nvidia attributes the increase to limited memory supply. It now sells for $6,950, nearly 75% above its launch price of about $4,000.
