NVIDIA’s New $4,999 DGX Spark Has Half the Memory — and Costs More Than the Original
What happened: NVIDIA announced a new 64GB configuration of its DGX Spark personal AI supercomputer on October 2, 2026, priced from $4,999 and available October 23. Why it matters: it has half the memory of the original 128GB model yet starts $1,000 above that machine’s original launch price — a stark illustration of how brutal memory costs have become in the AI era.
October 3, 2026 — Santa Clara.
What happened
NVIDIA unveiled a 64GB unified-memory configuration of the DGX Spark, its desk-side “personal AI supercomputer.” It keeps the same GB10 Grace Blackwell Superchip (a 20-core Arm CPU paired with a Blackwell GPU), 273 GB/s of memory bandwidth, DGX OS, and the full NVIDIA AI software stack — but halves the unified memory to 64GB of LPDDR5X.
The new configuration starts at $4,999 and will be sold exclusively through six hardware partners — Acer, ASUS, Dell, Gigabyte, HP, and MSI — beginning October 23. NVIDIA is not releasing a 64GB Founders Edition.
The details
- What it can run: NVIDIA says a single 64GB unit supports models with up to 100 billion parameters locally; two linked units pool 128GB and can handle up to 200 billion parameters. In one test (Qwen 3.8 27B), two clustered 64GB systems delivered up to 1.7× the performance of a single 128GB system.
- Networking: ConnectX-7 is carried over, letting users cable up to four units into a cluster.
- The price story: The original 128GB DGX Spark launched at $3,999. NVIDIA raised its MSRP to $4,699 in February 2026, citing worldwide memory supply constraints — and reports now put retail prices around $6,950. So the “cheaper” 64GB model at $4,999 still costs more than the 128GB machine’s original price.
Why it matters
This launch is less about a new product than about the memory market holding AI hardware hostage. Memory is the scarce resource: NVIDIA is charging $4,999 for 64GB of RAM attached to a great chip, and two of them (plus cabling) to do what one original box did for $3,999. If you’re a developer deciding between local and cloud inference, the math keeps shifting toward the cloud — unless you specifically need local control. For everyone else, the DGX Spark 64GB is a fascinating, expensive window into how the AI memory crunch is warping even NVIDIA’s pricing.
FAQ
Can the 64GB DGX Spark run large AI models? NVIDIA says yes: up to 100 billion parameters on one unit, or up to 200 billion with two units linked — enough for many open-weight models, though not the biggest ones.
Why does it cost more than the original 128GB model did? Memory prices have surged. NVIDIA raised the original model’s MSRP in February citing global memory constraints, and retail prices have climbed further since.
When and where can I buy it? October 23, 2026, through Acer, ASUS, Dell, Gigabyte, HP, and MSI.
Sources: NVIDIA, VideoCardz, The Register.