Nvidia confirms RTX Spark desktop sales from ASUS, Dell and MSI for local AI
The compact Windows machines target round-the-clock agents and creative work. Earlier MSI listings show the substantial upfront cost of that local computing capacity.
Loading page…
The compact Windows machines target round-the-clock agents and creative work. Earlier MSI listings show the substantial upfront cost of that local computing capacity.
Listen to this story
Nvidia’s November 18, 2026, announcement brings ASUS and Dell into the RTX Spark compact-desktop lineup alongside MSI, with systems from all three makers available to buy. The Windows PCs pair a 20-core Grace CPU with a Blackwell GPU and up to 128GB of unified memory for local model work and persistent agents. Nvidia also describes workflows that split tasks between local and cloud models, so these desktops offer an option for supported work—not a blanket replacement for cloud services. MSI’s listed configurations cost $4,699 and $7,999, not family-wide starting prices.
MSI’s EdgeMesa N AI+ listings paired 64GB of memory and 1TB of storage for $4,699, or 128GB and 4TB for $7,999; both specs change between configurations.
Nvidia claims support for local language models with 120 billion parameters, 3D scenes larger than 90GB, and real-time 4:2:2 video editing; these are not independent test results.
CUDA and Linux development tools through WSL support development, while Hermes Agent, NemoClaw blueprints, and Unsloth cover agent setup and supported-model fine-tuning.
Local AI work now has a dedicated desktop option from three PC makers. Nvidia said on November 18, 2026, that RTX Spark compact desktops from ASUS, Dell and MSI are available to buy. The Windows machines target agents, image and video generation, and model fine-tuning, with up to 128GB of unified memory.
Some sales activity was already underway before Nvidia’s announcement. In a November 14 article, TechRadar’s Rahim Amir reported that MSI’s EdgeMesa N AI+ had gone on sale in the United States that week. Its listed configurations paired 64GB of memory and 1TB of storage for $4,699, or 128GB and 4TB for $7,999.
Those prices describe two MSI configurations, not starting prices for the entire RTX Spark desktop family. They also change memory and storage together: the higher price is not simply the cost of adding more memory. TechRadar cited a November 11 Notebookcheck check that found the 128GB model out of stock, a dated observation rather than a current stock guarantee.
Nvidia’s November 18 announcement puts ASUS and Dell alongside MSI in the available-to-purchase group. The desktops join an existing RTX Spark laptop lineup, but the company gives the stationary machines a distinct job: keeping agents and scheduled tasks running while a laptop travels or handles other work.
The desktop specifications combine a 20-core Nvidia Grace CPU with a Blackwell RTX graphics processor containing 6,144 CUDA cores. Ten-gigabit Ethernet provides a wired connection for moving large datasets, model files and creative assets between the desktop and networked storage. Nvidia positions the machines as a single Windows system for development tools, creative applications and local models.
Nvidia says the hardware can run language models with 120 billion parameters locally. Its other stated workloads include rendering 3D scenes larger than 90GB and editing 4:2:2 video in real time. These are company-supplied capability claims, not independent test results. The memory specification is also a ceiling: Nvidia describes the family as offering up to 128GB.
Nvidia specifies up to 128GB for RTX Spark compact desktops; the earlier MSI listings included both 64GB and 128GB configurations.
For developers, the software foundation includes native CUDA support and Linux development tools through Windows Subsystem for Linux. Nvidia identifies Hermes Agent and its NemoClaw blueprints as routes to setting up agents, with Unsloth available for fine-tuning supported models. That puts model customization alongside running models, rather than limiting the desktop pitch to an always-on assistant.
Local execution has a specific cost and data benefit in Nvidia’s description: supported tasks processed on the PC keep their code and files there and avoid per-token cloud API charges for those tasks. That is narrower than saying every agent workflow stays offline. Nvidia also describes software that deliberately splits work between local and remote models.
Switchyard, available in Hermes, automatically routes agent tasks between the two. Nvidia says everyday tasks can run on the device while advanced reasoning draws on cloud models. For creative work, it describes Hermes connecting Photoshop and ComfyUI so natural-language instructions can guide a sketch into a rendered image or animation.
The next hardware wave remains pending. Nvidia lists compact desktops from Acer, Gigabyte and HP, plus Microsoft’s Surface RTX Spark Dev Box, as coming soon. For now, its available-to-buy desktop lineup names ASUS, Dell and MSI.
Loading discussion...
Join the conversation
Explain what would make the upfront cost worthwhile—or not.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.