AI · Sep 30, 2026
The four rules in the White House AI commitment, and the missing fifth: penaltiesA $4,999 64GB DGX Spark arrives October 23, with two-box clustering via Nvidia Sync
· Updated · Signal Desk newsroom
Analysis of reporting by NVIDIA Blog. How we work

Nvidia says a 64GB DGX Spark from Acer, ASUS, Dell, Gigabyte, HP and MSI ships October 23 from $4,999, using the same GB10 Grace Blackwell chip as the 128GB model. Two units can cluster over ConnectX-7, delivering up to 1.7x performance in Nvidia’s test.
Key takeaways
- The 64GB DGX Spark ships October 23 from $4,999 through six PC makers.
- It uses the GB10 Grace Blackwell Superchip, DGX OS and Nvidia’s AI stack.
- Nvidia pitches local agents running models up to roughly 100 billion parameters.
- Two units can pool memory to 128GB via ConnectX-7 and Sync Cluster Assistant.
- Nvidia reports up to 1.7x performance on a Qwen 3.8 27B test with two boxes.
What Nvidia announced
Nvidia says a 64GB DGX Spark configuration ships from Acer, ASUS, Dell, Gigabyte, HP and MSI starting Friday, October 23, with prices from $4,999 (technical details). It uses the same GB10 Grace Blackwell Superchip, DGX OS and Nvidia AI software stack as the 128GB model. The pitch is on-device agents running models up to roughly 100 billion parameters without relying on the cloud for every task.
Two units can be clustered over the built-in ConnectX-7 network interface using Nvidia Sync Cluster Assistant, pooling memory to 128GB. In Nvidia’s Qwen 3.8 27B test, two boxes delivered up to 1.7 times the performance of one. A Model Launcher due at month’s end promises one-click local launch across a cluster. DGX Spark sits alongside RTX Spark Windows PCs from major manufacturers, and Nvidia is pointing creators to open models such as Alibaba’s Qwen-Image-2.1 on its hardware.
Making private inference routine
The interesting move is less another GPU configuration than treating clustering as a near-consumer workflow: connect two boxes, let software configure the link and keep the same stack. That serves builders who want agents running overnight on their own data, fine-tuning experiments or longer contexts without spinning up cloud instances for every loop. Nvidia has been pushing local AI on several fronts (IFA local AI, local AI community).
For enterprises and labs, personal supercomputers through partner channels can sit beside regulated data, though a two-node DIY cluster is still operations work in desktop form. Each capable local box chips away at the habit of always reaching for an API during prototyping, even if training and peak demand stay in the cloud.
Caveats
$4,999 is accessible only by workstation standards, and partner exclusivity keeps Nvidia’s platform at the center of the chain. Local AI that works only within Nvidia’s operating system and agent toolkit is independence with a vendor’s name attached. The 1.7x figure is Nvidia’s own test.
Whether two-node Sync sees real use or remains a demo, and whether Blender support and agent playbooks land cleanly, will decide if this is a genuine alternative to renting tokens or a polished mini PC. More in our Nvidia coverage, tools coverage and agents coverage.
Who feels it
- Developers and researchers
- A lower-cost entry point for private agents and fine-tuning.
- Enterprises and labs
- On-premises boxes near regulated data, with ops overhead for clusters.
- Cloud providers
- Local boxes erode always-on API habits for prototyping.
What to watch
- Whether October 23 pricing and availability hold at $4,999.
- Sync Cluster Assistant reliability beyond Nvidia’s demo.
- Uptake of Model Launcher and Blender support.
FAQ
When does the 64GB DGX Spark ship?
October 23, from Acer, ASUS, Dell, Gigabyte, HP and MSI.
How much does it cost?
From $4,999.
What chip does it use?
The GB10 Grace Blackwell Superchip, like the 128GB model.
Can units be combined?
Yes, two can cluster over ConnectX-7 to pool 128GB of memory.
Companies: NVIDIA
More on NVIDIA
AI · Oct 7, 2026
Microsoft prices its first RTX Spark PCs; the agent sandbox coming to all Windows 11 users may matter moreAI · Sep 24, 2026
Nvidia, DeepMind and EMBL-EBI release predicted structures for protein complexes from 2,800-plus virusesAI · Oct 21, 2026
Europe’s banks are training their own AI on open weights, with Nvidia underneathAI · Oct 21, 2026
Nvidia turns Cosmos into a synthetic-data factory for robots’ rarest, riskiest moments