Nvidia DGX Spark 64GB Price: $4,999, the 128GB Jump to $6,950, and the Memory Shortage Behind Both

Nvidia's new 64GB DGX Spark holds a $4,999 price by cutting memory in half — and the reason it exists at all is that the original 128GB model it launched at $3,999 now costs $6,950, a roughly 74% jump driven entirely by the 2026 global memory shortage.

This is a small desktop AI workstation, not a data-center GPU cluster, but it's one of the cleanest public data points yet on how far the DRAM and HBM shortage has traveled: it's now reshaping the price and spec sheet of a $5,000 consumer-adjacent product, not just hyperscaler GPU contracts. We first flagged the launch in our Pulse coverage of the memory crunch; this piece goes deeper on the pricing math and the supply story behind it.

Nvidia DGX Spark 64GB: $4,999, Launching October 23

Nvidia is adding a 64GB unified-memory configuration of the DGX Spark, starting at $4,999 and sold through Acer, Asus, Dell, Gigabyte, HP, and MSI beginning October 23, 2026, Nvidia announced on its own blog on October 2, 2026, a launch Tom's Hardware also covered. The new SKU carries the same GB10 Grace Blackwell Superchip, ConnectX-7 networking, DGX OS, and Nvidia's AI software stack as the 128GB model, but a single 64GB unit tops out running models with up to 100 billion parameters, versus up to 200 billion parameters on the 128GB model.

Figures from Nvidia's October 2026 DGX Spark pricing as reported by Tom's Hardware and StorageReview.

Two 64GB units can be linked over the shared ConnectX-7 networking to pool memory to 128GB, the same approach StorageReview detailed in its coverage of the launch. That costs roughly $10,000 combined — about 44% more than a single $6,950 128GB unit — but buyers get double the compute and bandwidth for the difference, which matters for workloads that are compute-bound rather than purely memory-bound.

Why: The 2026 DRAM and HBM Shortage

The DGX Spark repricing is a symptom of a much larger supply crunch. The Register reported DRAM prices rising roughly 70% on AI-driven demand, and Gartner has projected a 130% year-over-year DRAM price increase across all of 2026. The root cause is straightforward: a single AI server reportedly requires 8 to 10 times more memory than a conventional server, and that demand is pulling wafer capacity away from the consumer and workstation-grade memory that products like the DGX Spark rely on.

Samsung, SK Hynix, and Micron are the three suppliers that matter most here, and all three have redirected capacity toward High Bandwidth Memory for AI data centers rather than the general-purpose DRAM and LPDDR5X that go into PCs, smartphones, and workstations like the Spark. The Register separately reported that Samsung and SK Hynix raised HBM3E supply prices by nearly 20% for 2026, with HBM effectively sold out for the year — leaving less fab capacity, at any price, for everything else.

Micron has projected the supply-demand imbalance could persist through 2027 and into 2028. That matters for anyone buying AI hardware in the next two years: this isn't a one-quarter spike that corrects itself, it's a structural reallocation of global memory supply toward AI data centers that leaves less for everything built on the same memory types.

What the Headline Misses

The "Nvidia raised prices" framing undersells how unusual this response actually is. Nvidia didn't just raise the 128GB DGX Spark's price and move on — it built an entirely new, lower-memory SKU specifically to hold a sub-$5,000 entry point for local AI developers who don't need 128GB. That's a company absorbing real supply-cost pressure by segmenting its product line rather than simply passing the full cost increase on to every buyer.

The honest caveat: a 64GB unified-memory machine is a real downgrade for anyone running larger local models, not just a repackaged discount. Buyers choosing the 64GB tier to save money need workloads that actually fit in half the memory, or they're trading capability for a price point that still costs as much as a well-specced gaming PC.