Search  for anything...
msi

msi EdgeXpert AI Mini Desktop (DGX Spark Platform), NVIDIA GB10 Grace Blackwell, 128GB LPDDR5 Unified Memory, 4TB NVMe Gen5 SSD, WiFi 7, BT 5.3, NVIDIA DGX OS (Linux): 13SUS Black

  • Based on 0 reviews
Condition: New
Checking for the best price...
$4,699.00 Why this price?

Buy Now, Pay Later

As low as / mo
  • – Up to 36-month term if approved
  • – No impact on credit to apply
  • – Instant approval decision
  • – Secure and straightforward checkout

Ready to go? Add this product to your cart and select a plan during checkout.

Payment plans are offered through our trusted finance partners Klarna, Affirm, Afterpay, Apple Pay, and PayTomorrow. No-credit-needed leasing options through Acima may also be available at checkout.

Learn more about financing & leasing here.

Selected Option

Free shipping on this product

FREE 30-day refund/replacement

To qualify for a full refund, items must be returned in their original, unused condition. If an item is returned in a used, damaged, or materially different state, you may be granted a partial refund.

To initiate a return, please visit our Returns Center.

View our full returns policy here.

Availability: In Stock.
Fulfilled by Amazon
Arrives Sunday, Jul 19
Order within 3 hours and 42 minutes
Available payment plans shown during checkout
Protection Plan Protect Your Purchase
Checking for protection plans...

Recently Viewed

Size: 4TB NVMe gen5 SSD


Style: NVIDIA GB10 Grace Blackwell


Features

  • AI Performance: Run Large AI Models Locally Powered by NVIDIA GB10 Grace Blackwell architecture, delivering up to 1000 TOPS of AI performance for generative AI, LLMs, and advanced edge computing workloads.
  • CPU: High-Performance Arm CPU Architecture 20-core design with high-performance and efficiency cores enables smooth multitasking, faster data processing, and optimized power usage for demanding AI applications.
  • Memory: Massive 128GB Unified Memory LPDDR5X high-bandwidth memory (up to 273 GB/s) allows efficient handling of large datasets and AI models without bottlenecks, support large-scale AI models up to 200 Billion Parameters.
  • Storage: Ultra-Fast 4TB Gen5 SSD Storage Experience lightning-fast load times and data access with PCIe Gen5 NVMe SSD (up to 10,000 MB/s), plus self-encrypting capabilities for enhanced data security.
  • Connectivity: Next-Gen Connectivity for Edge AI Equipped with WiFi 7, Bluetooth 5.3, USB4 Type-C, and high-speed networking options including ConnectX-7 for low-latency, high-bandwidth environments.
  • OS / Use Case: Optimized for AI Development Pre-installed NVIDIA DGX OS (Ubuntu Linux-based), tuned for machine learning, AI model training, and edge deploymentideal for developers, researchers, and enterprise AI applications.

Description

MSI EdgeXpert is a compact AI supercomputing desktop built on the NVIDIA DGX platform, designed to bring powerful AI capabilities directly to your workspace. Equipped with NVIDIA GB10 Grace Blackwell architecture, 128GB of high-bandwidth unified memory, and ultra-fast Gen5 storage, it enables developers and professionals to run large AI models locally—reducing reliance on cloud infrastructure while improving speed, privacy, and cost efficiency. Its ultra-compact form factor makes it ideal for edge deployments, research labs, and enterprise environments where space and performance matter. With next-generation connectivity including WiFi 7 and high-speed networking, EdgeXpert seamlessly integrates into modern AI workflows. Whether you're developing generative AI, running inference at the edge, or prototyping advanced models, MSI EdgeXpert delivers data center-class performance in a desktop-sized solution.

Brand: msi


Operating System: NVIDIA DGX OS


CPU Model: ARM Cortex A5


CPU Speed: 3.25 GHz


Cache Size: 32 MB


Graphics Card Description: Dedicated


Graphics Coprocessor: NVIDIA Blackwell Architecture


Memory Storage Capacity: 4 TB


Specific Uses For Product: Business, Everyday Use, Multimedia


Personal computer design type: Mini PC


Operating System: NVIDIA DGX OS


Specific Uses For Product: Business, Everyday Use, Multimedia


Personal Computer Design Type: Mini PC


Color: Black


Additional Features: SSD, Wi-Fi 7


Hard Disk Description: SSD


Hardware Interface: Ethernet, HDMI, USB Type C


Power Consumption: 240 Watts


Item Dimensions: 5.94 x 5.94 x 2.05 inches


Item Weight: 2.7 Pounds


Video Output Interface: HDMI


Hard Disk Interface: Solid State


Style Name: NVIDIA GB10 Grace Blackwell


Compatible Devices: Keyboard, Mouse, Monitor, Television, Speaker, Printer, External Storage Drive, Game Controller


Power Plug Type: Type C


Video Output: HDMI


Cache Memory Installed Size: 32 MB


Memory Storage Capacity: 4 TB


RAM Memory Installed: 128 GB


RAM Memory Technology: DDR5


Ram Memory Maximum Size: 128 GB


Memory Speed: 6400 MT/s


RAM Type: DDR5 RAM


Memory Clock Speed: 1067 MHz


Display Resolution Maximum: 1920x1080


Aspect Ratio: 169


Resolution: 1920 x 1080


Native Resolution: 1920 x 1080


Connectivity Technology: Bluetooth, Ethernet, HDMI, USB


Wireless Compability: 802.11ax, Bluetooth


Wireless Technology: Bluetooth, Wi-Fi


Brand: msi


Model Number: EdgeXpert13S


Model Name: EdgeXpert-13SUS


Built-In Media: Desktop


Processor Brand: 20 core ARM


Model Year: 2025


CPU Model Number: 10 Cortex-X925 + 10 Cortex-A725 Arm


Warranty Description: 1 Year Manufacturer


Video Processor: NVIDIA


UPC: 824142447444


Unit Count: 1.0 Count


Manufacturer: MSI


Graphics Description: Dedicated


Graphics Coprocessor: NVIDIA Blackwell Architecture


Graphics Card Ram: 128 GB


Graphics Ram Type: DDR5 SDRAM


Graphics Card Interface: PCI Express


Processor Series: ARM Cortex A5


Processor Speed: 3.25 GHz


Processor Socket: BGA 1023


Processor Count: 1


CPU Model Speed Maximum: 3.8 GHz


Total Usb Ports: 4


Total Number of HDMI Ports: 1


Number of Component Outputs: 1


Human-Interface Input: Keyboard, Mouse


Keyboard Description: Standard QWERTY Wired Keyboard


Keyboard Layout: QWERTY


Frequently asked questions

If you place your order now, the estimated arrival date for this product is: Sunday, Jul 19

Yes, absolutely! You may return this product for a full refund within 30 days of receiving it.

To initiate a return, please visit our Returns Center.

View our full returns policy here.

  • Klarna Financing
  • Affirm Pay in 4
  • Affirm Financing
  • Afterpay Financing
  • PayTomorrow Financing
  • Financing through Apple Pay
Leasing options through Acima may also be available during checkout.

Learn more about financing & leasing here.

Similar Products

Top Amazon Reviews

    • Unified RAM with tradeoff of some speed
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
MSI's version of the DGX Spark is definitely one of the best. I've seen others complain about overheating and shutdowns, but I've had none of that. Despite that, I decided to put a Wathai 140mm x 25mm fan underneath for extra airflow. Under concurrent inference load, it stays under 80C and performs well. Upon receiving it, I noticed significant disk usage. The manufacturer left disk burn-in files, totaling a few TB, on the device, but once I found them, deleting them resolved that. The DGX platform itself is a mixed bag. Nvidia has been slow to push through better support for the device in libraries like vLLM and llama.cpp. The best performing setup I've used for coding agents so far has been using Intel's qwen3.5-122b-a10-int4-autoround. With that I get generation speeds of 30 t/s, and upwards of 65 t/s under currency=4. Prompt processing speeds are not an issue, usually well into the thousands of t/s. Overall, it's a solid device which handles concurrency really well. Although, to be real, what you're really getting is the 128GB of unified RAM at the cost of slower bandwidth. If you want raw speed, and are willing to sacrifice a little memory and concurrency, go with a Blackwell card. I am hopeful though that libraries, and quants like NVFP4, will be able to maximize performance ... show more
Reviewed in the United States on April 25, 2026 by B. Jester
    • Great performance
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
Great performance! NVIDIA and Msi are bringing the best AI hardware for the price in a small package.
Reviewed in the United States on April 13, 2026 by MarylandAmazonFan
    • More impressive than I initially thought
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
This is a review of the DGX Spark / GB10 platform in general, not MSI's packaging of it. I am updating my original 1-2 star review and bumping it up to 4. After a month of working with it nearly around the clock, I have found some ways to make it do useful things that my Strix Halo Max+ 395 cannot. Performance of GGUFs through llama.cpp on the GB10 is not much of an improvement over when you'll see on Max+ 395 device, which costs half as much, though it does perform better with the large context prompt processing, just not the inferencing. tensorrt-llm simply is not ready for the GB10 yet, and it's difficult or impossible to get that engine to do anything interesting at all. But vLLM with a properly tuned nvfp4 can be most impressive. A lot of work is required to get a 70B llama 3.3 model into a nvfp4 format optimized for vLLM. Converting from bf16 to nvfp4 cannot be done on the device itself and requires 1-2 hours of B200 or 2xH100 rental time, since the .safetensors will not fit into 128GB. That step will cost about $10. The rest can be done directly on the GB10. This is not the forum for writing instructions on how to go about that, but with a few scripts that Claude or any AI can help you with, you can get appropriate k/v weights into the model, fix some data types, and wind up with a 70B llama 3.3 nvfp4 quant that can handle up to the full supported 128K token context with speeds above 5t/s. I'm getting that speed at about 30K in use. I imagine it will degrade some with additional growth, but this is still better than llama.cpp can do and infinitely better than the Max+ 395 can do. vLLM for GB10 has a special docker container for the 119B nvfp4 Mistral 4 Small MoE model that runs so fast it makes your head spin! 30-40t/s from a 119B on a box you can run in your house is pretty impressive indeed! The Mistral model is not tuned for my use case, but with a little training it might be. With all that said, I STILL think the GB10 is a bit overhyped and not ready for every use case. For this reason, I cannot give it 5 stars. It is immature and the software landscape is in such a state of flux that every document you find was obsolete a week before you read it. But if you like that sort of thing ... have at it! The GB10 does have enormous potential, and it can perform impressively even today. I've been working on this review for a while. My initial one gave it 1 star; I was so furious with my inability to get it to do *anything* Nvidia claimed it could do that I opened a return request. Then I discovered that my Max+ 395 was not able to process the prompt context I had generated on the GB10 without causing a timeout that I could not resolve. So I bumped the review to 2 stars and delayed my return. Then I got vLLM to do these legitimately impressive things, and I had to revise my review altogether, and I've decided to keep the device. I apologize to any who factored my earlier, more scathing, reviews into their decision about whether or not to buy this product. I promise that these are my final words on the topic. ... show more
Reviewed in the United States on June 4, 2026 by Rob M.
    • Not a Frontier-LLM Competitor, but... Not a Frontier-LLM Competitor, but...
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
Review Image by Haruyuki Shini
This computer is capable of running NVIDIA's Nemotron 3 Super model LLM, which punches way above its weight class. You can easily get a 400k context window while also running Stable Diffusion and generating images that are 1080p or bigger. If you make it a dedicated LLM machine, you can easily get over a 1 million token context window. The only thing I don't like about this machine is the Ubuntu-default stuff like GNOME (ugh!), but Linux is Linux. You can really set this up however you like. I recommend setting it up and then using SystemD to disable gdm for even more RAM headroom (do not do this unless you know *exactly* what that means). It's also pretty quiet. Running both the LLM and Stable Diffusion at the same time pushed it into somewhat uncomfortable thermal territory, but running just one at once tends to stay in the 70s-degree range. The WiFi is good too. Usually I'm a WiFi hater, but honestly the amount of network data throughput this machine needs is rather low. It works great while only being connected to power. ... show more
Reviewed in the United States on March 25, 2026 Reviewed in the United States on March 25, 2026 by Haruyuki Shini
    • Great!
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
Amazing machine!
Reviewed in the United States on May 23, 2026 by Roman H
    • Very happy
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
Arrived on time in great shape. Works great.
Reviewed in the United States on April 28, 2026 by Chris JB Gutierrez
    • A great tiny AI machine
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
it's a great little machine. Running my models helping me optimize some applications.
Reviewed in the United States on February 20, 2026 by Skender Kajoshaj
    • Útil pero requiere trabajo
Size: 4TB NVMe gen5 SSD Style: NVIDIA GB10 Grace Blackwell
Cumple con lo esperado y es definitivamente útil para poder trabajar con modelos de LLMs más pesados (fine-tuning, continual pre-training, entre otras técnicas). Sin embargo, al ser el chip bastante nuevo/novedoso no tiene buen soporte con librerías de Python y cuesta un viaje resolver incompatibilidades para que realmente se le saque provecho a este hardware. Para inferencia, es definitivamente más lento que la línea de mayor VRAM RTX. ... show more
Reviewed in the United States on May 21, 2026 by Javier

Can't find a product?

Find it on Amazon first, then paste the link below.
Checking for best price...