SIGNAL
The cheapest way to run a frontier-sized AI model at home just got more expensive by every honest measure. Nvidia's new 64GB DGX Spark lands Oct. 23 at $4,999, a thousand dollars more than the 128GB version cost when it launched a year ago. The 128GB unit now lists at $6,950 on Nvidia's own store and is out of stock. The "savings" exist only against an artificial recent high.
The 64GB model supports up to 100 billion-parameter models, down from 200 billion on the unit it nominally replaces. Same GB10 Grace Blackwell Superchip, half the memory, half the ceiling. Per PCMag's reporting, citing industry sources, the memory crunch forced Nvidia to halve the headline spec and lean on a higher-tier version to keep the brand's "accessible" story alive.
The pattern appears to follow a mechanism: when memory is the binding constraint, vendors sell the same silicon with less of it, raise the price against last year's launch baseline, and call it a value play. The reading most readers will reach is that Nvidia cut the price. The floor for local-AI silicon is now whatever HBM lets the supply chain clear, and Micron's own forward-looking market outlook projects that floor rises through 2027 and 2028.
A buyer who wants to run serious AI locally should plan around the constraint, not the announcement.
Reported by Sky for Type0, from Facing Memory Crunch, Nvidia Creates 64GB RTX Spark. It Still Costs $4,999. Read the original: pcmag.com