Nvidia Raises DGX Spark 128GB to $6,950 and Adds a $4,999 64GB Model as Memory Costs Bite
Nvidia will launch a 64GB DGX Spark at $4,999 on 23 October while lifting the 128GB model to about $6,950 from its $3,999 launch price, a direct pass-through of sharply higher memory costs.
Nvidia is repricing the entry point for local AI development hardware. The DGX Spark 128GB/4TB model, which launched at about $3,999 roughly a year ago, is moving to around $6,950, with individual OEMs setting their own price points. To keep a lower rung on the ladder, Nvidia is adding a 64GB DGX Spark at $4,999, available from Acer, ASUS, Dell, Gigabyte, HP and MSI starting Friday 23 October, and it retains the ConnectX-7 200GbE networking that ships in the 128GB version so two units can be clustered for roughly $8,000.
The repricing is a direct consequence of memory costs rather than a change in silicon. ServeTheHome's editor-in-chief noted the lab is now paying about $2,000 for 64GB ECC RDIMMs, so system pricing logically has to rise, and the site built an eight-node GB10 cluster earlier this year at about $3,000 per node, a configuration that would cost more than $50,000 today. Framed per gigabyte of memory, the 128GB box has gone from roughly $31 per gigabyte at launch to about $54 at the new list price.
The new pricing also changes the competitive set. At $6,950 the DGX Spark overlaps AMD Gorgon Halo systems arriving in October that offer about 50 percent more memory capacity at 192GB, and Apple's $10,000 256GB Mac Studio M5 Ultra is positioned just above it, so buyers are weighing memory capacity against Nvidia's networking and software stack. GB300-based DGX Station systems still sell for roughly $100,000, but the gap between a small GB10 cluster and a single large box is narrowing as the low end reprices.
Rack-level integration is where the volume is moving. Chenming Electronic Tech (UNEEC) president Charles Lo said L11 rack system integration accounted for 18 percent of revenue in January-August 2026, with group revenue of NT$7.9 billion, and that new models such as VR200 entering mass production in the fourth quarter should help monthly revenue recover toward its former level of more than NT$1 billion. That mirrors the wider trend of ODMs shifting from board-level to full-rack shipments as AI server platforms become the unit of delivery.