Nvidia DGX Spark 64GB Desktop Arrives at $4,999

Nvidia hits retail shelves on October 23, 2026, with a new $4,999 desktop. The Nvidia DGX Spark 64GB gives engineers local hardware for AI work. It pairs a Grace Blackwell Superchip with DGX OS in a compact chassis. You can put it right on your desk.

The launch comes right after a huge price hike on the 128GB edition. That bigger system cost $3,999 a year ago. Rising memory costs pushed its price tag to $6,950. The Register reported that the new unit gives you half the memory for $1,000 more than the original model's starting price.

Local compute saves money on cloud bills. It keeps your code in your office. It is quiet. However, you pay far more per gigabyte today than you did twelve months ago.

What you get inside the smaller box

The 64GB model shares key silicon with its predecessor. You get the same GB10 processor and unified memory. It includes the standard Nvidia AI software stack. Nvidia claims one box handles fine-tuning for models up to 100 billion parameters.

That capacity covers everyday engineering work. You can evaluate open weights without renting cloud servers. You can run private software agents. Your code stays off outside networks. That privacy matters to small teams.

High-speed networking comes standard on this machine. Every unit includes ConnectX-7 200GbE ports. Data moves fast. If your work fits inside 64GB of memory, the machine runs smoothly. It works right out of the box.

High memory costs are driving up prices

High memory prices changed Nvidia's hardware plan. A global squeeze on fast RAM hurt supply chains across the industry. PCMag reported that memory shortages pushed the existing 128GB system near $7,000. That left an empty spot for cheaper hardware.

Nvidia cut the memory to keep entry prices under five figures. Hardware partners will build and sell these desktop units. The list includes Acer, ASUS, Dell, Gigabyte, HP, and MSI. Buyers still face a steep price per gigabyte compared to last year.

Supply remains thin across retail channels. ServeTheHome reported that the original 128GB Founders Edition showed as out of stock on Nvidia's U.S. store. Hardware makers are adjusting factory production lines. Fast memory supply cannot meet current market demand.

Linking two units doubles your capacity

You can link two boxes together if you need more memory. The built-in ConnectX-7 port joins two systems into one compute pool. Nvidia says its software recognizes both units automatically. That setup creates 128GB of pooled memory.

Two linked units support models up to 200 billion parameters. You can run large open models without cloud accounts. Your team can buy one machine today. You can add a second unit later when project budgets grow.

The total cost of linking two boxes adds up quickly. Buying two 64GB units at $4,999 each brings your bill to roughly $10,000. That costs much more than the $6,950 sticker price of a single 128GB model. However, you must actually find the 128GB version in stock.

Six hardware partners will build the desktop

Nvidia is not selling this system alone. Distribution depends on third-party PC builders. They package the GB10 chip inside their own chassis designs. Stores will stock these machines starting October 23, 2026.

Base models start at $4,999. Storage options like 1TB or 4TB solid-state drives will alter final vendor tags. You will see these workstations appear in catalogs from six primary manufacturing partners.

We expect varied chassis shapes from these vendors. Some may offer extra cooling options. Others might target quiet home office setups.

• Acer
• ASUS
• Dell
• Gigabyte
• HP
• MSI

How you should weigh this purchase

First, check the size of the models you run every week. If you use models under 70 billion parameters, 64GB of memory does the job. You avoid cloud latency. You get private local compute without monthly bills.

If you routinely run models over 100 billion parameters, this machine will feel small. Linking two units costs $10,000. That price tag might shock a small startup. Finding a standalone 128GB unit at $6,950 makes better financial sense for larger builds.

Audit your monthly cloud bills before October 23 arrives. Calculate how many months of cloud compute $4,999 actually covers for your setup. If your cloud spend tops that number within eighteen months, buying the desktop makes sense despite the higher memory tax. You gain permanent hardware on your desk.