What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Yes. As of August 18, 2026, NVIDIA’s U.S. Marketplace lists DGX Spark for $4,699 with an Add to Cart option, while Best Buy lists a PNY-delivered unit for $5,299.99. That makes the system orderable, not universally in stock: delivery estimates depend on location, and partner logos do not confirm local inventory. “Hard to find” is best understood as uneven distribution and price variation—not proof of a nationwide shortage.
Where can you buy DGX Spark, and what does it cost?
These are U.S. listing observations from August 18, 2026. An Add to Cart button means a buyer can attempt to place an order; it does not guarantee immediate shipment to every address.
| Listing | Observed price | Availability signal | What to know |
|---|---|---|---|
| NVIDIA Marketplace, single DGX Spark | $4,699 | Add to Cart | NVIDIA’s first-party U.S. listing. Check the page for current terms. |
| Best Buy listing, PNY-delivered single unit | $5,299.99 | Add to Cart; shipping and pickup estimates displayed | At the selected Roseville location, shipping was estimated for August 20 and pickup for August 21, 2026. Those dates are location-specific, not a national stock statement. |
| Best Buy two-unit kit | $9,999.99 | Add to Cart in a result crawled about three weeks before August 18, 2026 | This is an older price and availability signal; recheck the listing before making a decision. |
The two current single-unit listings differ by $600. The available information does not establish why; compare seller, warranty, shipping, tax, and return terms rather than assuming the higher price buys a different configuration. NVIDIA’s Marketplace page identifies Amazon, Micro Center, and PNY as authorized channel partners, but those partner indications do not confirm stock at every retailer or location. NVIDIA also announced expanded partner availability through Acer, ASUS, Dell, GIGABYTE, HP, Lenovo, and MSI beginning in July 2026; an OEM system may differ from NVIDIA-branded DGX Spark in operating system, storage, support, warranty, and chassis. Check the exact model and configuration with the seller. NVIDIA’s partner announcement
Why “hard to find” is only partly accurate
NVIDIA’s Marketplace listing and product page establish that DGX Spark is being sold and shipped through NVIDIA and partners. Best Buy’s orderable listing is another concrete retail signal. Together, they contradict the idea that the system is still merely announced or preorder-only. They do not establish that every retailer has units ready to ship, or that the official price is available in every region.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
- Distribution runs through a relatively focused group of specialist and authorized channels rather than broad, uniform retail shelves.
- Retailer inventory and delivery estimates can change by ZIP code and store.
- A GB10-based OEM system may be related to DGX Spark without being the same branded product or configuration.
- Marketplace listings can have different sellers and prices. A product page alone does not establish who fulfills an order or what warranty applies.
The careful verdict is that DGX Spark is available to order in the United States, but buyers may not find the exact configuration at NVIDIA’s listed price or with immediate local delivery.
What you get for the price
DGX Spark is a compact desktop AI development system built around NVIDIA’s GB10 Grace Blackwell Superchip. It is not a conventional GeForce gaming PC, nor is it simply a desktop with a discrete RTX card. NVIDIA lists DGX OS and an NVIDIA AI software stack, positioning the system for local AI development and inference. Its standard configuration includes a 20-core Arm CPU, Blackwell GPU architecture, 128GB of coherent unified LPDDR5x memory, and a 4TB self-encrypting NVMe SSD. It also lists ConnectX-7 networking, 10Gb Ethernet, Wi-Fi 7, Bluetooth 5.4, and a 240W power supply. The enclosure measures 150 × 150 × 50.5mm and weighs about 1.2kg. NVIDIA’s product specifications and DGX Spark documentation
What NVIDIA’s “1 PFLOP” claim means
NVIDIA describes performance of up to 1 PFLOP for theoretical FP4 AI computation under its stated conditions and sparsity assumptions. That figure is not a general CPU or GPU benchmark, and it does not tell you gaming performance, application speed, or tokens per second. Real results depend on the model, runtime, precision, context length, software support, and workload. Treat the claim as a description of a specific theoretical compute metric, not a promise of a particular result for your project.
Rank #2
- 900-5G172-2260-000
What the software offer includes
NVIDIA’s Marketplace listing advertises a free 90-day NVIDIA AI Enterprise license and a complimentary NVIDIA Deep Learning Institute hands-on course described as a $90 value. The license is a limited promotional entitlement, not a permanent subscription. Confirm the current listing and terms, and distinguish included software from any later subscription or support decision. Retailer-specific financing, warranties, protection plans, and setup services are separate terms to verify with that retailer.
Will it run the models and tools you need?
DGX Spark is aimed at local LLM inference, model prototyping and fine-tuning, agent development, robotics and edge-AI work, computer vision, and projects where keeping data on the local machine matters. NVIDIA highlights tools and workflows including PyTorch, Jupyter, Ollama, NVIDIA NIM, and NVIDIA Blueprints. That positioning is not a guarantee that every model, container, or binary will work unchanged.
The system’s 128GB unified memory is useful capacity, but it is not interchangeable with 128GB of discrete GPU VRAM. Actual model fit and speed depend on quantization format, architecture, context length, runtime overhead, and the KV cache used during inference. A model loading successfully is not the same as generating tokens quickly enough for your intended use. Verify that the specific frameworks, containers, drivers, and development tools your workflow requires support DGX OS and its Arm environment. Buyers accustomed to x86 Linux or Windows should pay particular attention to compatibility before ordering.
Rank #3
- Extreme AI Performance: Powered by NVIDIA GB10 Grace Blackwell Superchip delivering 1 petaFLOP of AI performance and 128GB memory for 200B model fine-tuning.
- Developer-Optimized Platform: Designed for AI developers building secure, long-running agentic workflows, with compatibility across frameworks such as OpenClaw and NemoClaw, supporting private on-device inference, sandboxed execution, and governed data access.
- Scalable Architecture: Featuring NVIDIA NVLink-C2C for ultra-fast CPU-GPU memory communication and NVIDIA ConnectX-7 networking to support dual GX10 system stacking, unlocking superior scalability and performance.
- Advanced Thermal Design: Engineered cooling ensures sustained high performance and reliability in an ultra-small form factor.
- Full Stack AI Solution: The GB10 and NVIDIA AI software stack provide a full stack solution for AI development and deployment.
Is the $4,699 price sensible for your use?
The financial case rests on the combination of local memory capacity, NVIDIA’s software environment, and a small preconfigured system—not on a claim that it is the cheapest way to obtain computing power. It may be worth considering if you have a sustained need to develop or run supported AI workloads locally, value privacy, low-latency access, or offline operation, and prefer an integrated setup to building and maintaining a workstation.
- Consider it if your models and tools are supported on Arm and DGX OS, you specifically need a large-memory local AI platform, and you can justify a $4,699-plus purchase for that work.
- Compare a conventional workstation if you need x86 compatibility, Windows as your primary OS, replaceable GPUs, more conventional serviceability, gaming, video editing, or broad general-desktop value.
- Consider cloud rental if local AI workloads are occasional rather than sustained. Compare total usage costs and data-handling requirements for your own work; there is no universal point at which cloud or ownership is cheaper.
- Compare OEM GB10 systems on their exact terms rather than assuming they are identical to DGX Spark. Operating system, SSD, networking, software bundle, warranty, support, and price may differ.
DGX Spark is not marketed as a gaming-first system, and its theoretical FP4 figure is not a substitute for workload-specific benchmarks. Treat memory and storage as the listed configuration unless the seller or documentation confirms that user upgrades are supported. A display, keyboard, network connection, and suitable cabling may also be needed; check the box contents for the exact listing.
Recommended Free Tools
Should you buy one system or the two-unit kit?
Two DGX Spark systems can be linked, and the platform includes ConnectX-7 networking. A second unit can help with demanding workloads when the software and parallelization strategy can use it. It does not automatically turn the setup into one seamless 256GB memory pool or double the speed of every application. Model parallelism, networking configuration, and application support determine whether the extra system produces useful gains. The two-unit listing observed at $9,999.99 is not a current-price guarantee, so confirm both price and configuration before treating it as a purchase option.
What to check before placing an order
- Start with the NVIDIA Marketplace listing and compare it with the retailer’s current page.
- Confirm the exact model, configuration, and seller of record. Do not infer that a marketplace seller is NVIDIA or an authorized partner solely from a similar product title.
- Enter your ZIP code or select your store before relying on delivery or pickup dates; treat an Add to Cart button as orderability, not guaranteed immediate shipment.
- Compare the total checkout cost, including any applicable tax and shipping, and read the seller’s return window and warranty terms.
- Check that your required frameworks, containers, binaries, and drivers support the Arm environment and DGX OS.
- Estimate whether your target models fit in the available memory after accounting for the operating system, runtime, context length, and KV-cache requirements.
- Confirm the included software, license duration, box contents, display connections, and network requirements for the exact listing.
- If considering two units, verify that your application supports multi-system execution and that the expected benefit justifies the added cost.
NVIDIA reported raising the Founders Edition price from its original $3,999 positioning to $4,699 in February 2026, citing constrained worldwide memory supplies. That historical price explains why older articles may quote $3,999; it is not the current NVIDIA Marketplace price observed in August 2026. Tom’s Hardware report on the February 2026 price change
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
