Nvidia announced a 64GB version of its DGX Spark desktop AI system Friday, starting at $4,999. The new entry model costs $1,000 more than the 128GB original did at launch, with half the memory. Developers now face a higher starting price and a smaller memory pool for local AI work.
Nvidia raised the price of the 128GB DGX Spark to $6,950. Nvidia’s own 128GB unit remains out of stock, while available partner systems were selling for roughly $7,000 to $9,000 at the announcement. The smaller configuration goes on sale Oct. 23 exclusively through Acer, Asus, Dell, Gigabyte, HP and MSI. Nvidia will not sell a branded Founders Edition of that model.
What Changed
- Nvidia's 64GB DGX Spark starts at $4,999 and goes on sale Oct. 23, only through Acer, Asus, Dell, Gigabyte, HP and MSI.
- The new entry model costs $1,000 more than the 128GB DGX Spark did at its October 2025 launch, with half the memory.
- The 128GB model now costs $6,950, up from $3,999 at launch and $4,699 after a February increase.
- Two clustered 64GB units cost $9,998, about 44% more than one 128GB system, while an AMD-based GMKtec mini PC with 128GB costs $3,650.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
The price ladder
The 128GB system launched in October 2025 at $3,999. Nvidia raised its price to $4,699 in February 2026, citing memory supply constraints. That increase was $700, about 18%. The October increase adds $2,251, about 48%, more than three times the February rise in dollars. Its latest price is nearly 75% above launch.
Nvidia has not published the smaller model’s storage capacity. Partners set final prices around the SSD they ship, making $4,999 a starting figure.
What the smaller model keeps
The new configuration retains the GB10 Grace Blackwell Superchip and its 20-core Arm CPU, co-designed with MediaTek. Memory bandwidth stays at 273 GB/s. ConnectX-7 networking, DGX OS and Nvidia’s AI software stack also carry over.
FREE AI BRIEFING · WEEKDAYS
Keep tabs on what local AI hardware costs.
Get the AI stories shaping the day, with concise context from San Francisco. The briefing takes about five minutes and arrives at 4:45 a.m. Pacific, 7:45 a.m. Eastern.
Free. No hype. Unsubscribe anytime.
Nvidia says a single 64GB system supports models up to 100 billion parameters, against up to 200 billion on the 128GB model.
Nvidia’s argument is that open models in the 26 billion to 35 billion parameter range, including Qwen 3.8 27B, are now capable enough for local coding and research agents. The smaller memory pool leaves less room for fine-tuning.
The clustering pitch
Two units connect directly by QSFP cable over a 200GbE ConnectX-7 link. Nvidia says the pair pools memory to 128GB, supports models up to 200 billion parameters and doubles memory bandwidth.
In Nvidia’s Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7 times the performance of a single system. That is Nvidia’s own measurement. There is no independent measurement yet.
At the announced starting price, two units cost $9,998 for the same memory capacity a single $6,950 system offers. That is $3,048, or about 44%, more. The $9,998 figure is a floor because partners price around the SSD they ship. The pair supplies twice the compute and bandwidth.
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
Nvidia Sync Cluster Assistant detects connected units, validates their configuration and sets up the network. Sync Model Launcher, due at the end of October, will launch Qwen 3.8 27B on one unit or a cluster.
Cheaper alternatives and supply
At the announcement, GMKtec’s Ryzen AI Max+ 395 mini PC cost $3,650 with 128GB and $2,349 with 64GB. The 128GB price is roughly half of Nvidia’s $6,950. AMD’s Gorgon Halo chips support up to 192GB. GB10’s GPU has been faster than AMD’s chips on AI work in published reviews.
Micron expects the memory crunch to worsen in 2027 and 2028, despite efforts to expand manufacturing capacity.
“We do not have line-of-sight to when supply and demand will return to balance,” Micron said.
Frequently Asked Questions
What is different about the 64GB DGX Spark?
It has half the unified memory of the original. It keeps the GB10 Grace Blackwell Superchip, the 20-core Arm CPU co-designed with MediaTek, 273 GB/s of memory bandwidth, ConnectX-7 networking and DGX OS. Nvidia says one 64GB unit supports models up to 100 billion parameters, against up to 200 billion on the 128GB model.
How much has the 128GB DGX Spark's price risen?
It launched in October 2025 at $3,999. Nvidia raised it to $4,699 in February 2026, citing memory supply constraints, and it now costs $6,950, nearly 75% above launch. Nvidia's own unit is out of stock, and available partner systems were selling for roughly $7,000 to $9,000.
Is clustering two 64GB units cheaper than buying one 128GB system?
No. Two units at the $4,999 starting price cost $9,998, about $3,048 more than one $6,950 system for the same 128GB of memory. The pair supplies twice the compute and bandwidth. Nvidia's own test showed up to 1.7 times the performance of a single system, and there is no independent measurement yet.
When and where can buyers get the 64GB model?
It goes on sale Oct. 23, exclusively through Acer, Asus, Dell, Gigabyte, HP and MSI. Nvidia will not sell a Founders Edition of it. Nvidia has not published the storage capacity, and partners set final prices around the SSD they ship, so $4,999 is a starting figure.
Will memory prices come down soon?
Micron expects the memory crunch to worsen in 2027 and 2028 despite efforts to expand manufacturing capacity. The company said it does not have line-of-sight to when supply and demand will return to balance.
AI-generated summary, reviewed by an editor. More on our AI guidelines.


Free AI briefing · Weekdays
The AI stories that matter, sourced and explained.
Join the Morning Briefing. It goes out every weekday at 4:45 a.m. Pacific, with later sends for the East Coast, Berlin and Tokyo.
Free when you sign up: The Paperclip Compendium, our tested guide to running AI agents.
Free. Unsubscribe in one click.
IMPLICATOR