NVIDIA launches RTX Spark small desktops from ASUS, Dell and MSI
NVIDIA’s RTX Spark small desktops have launched from ASUS, Dell and MSI. The dedicated Windows PCs support local AI agents, image and video generation, and model fine-tuning. They offer up to 128GB of unified memory. Pricing and individual configurations were not included in the announcement.
Key facts
- RTX Spark small desktops have launched from ASUS, Dell and MSI.
- The dedicated Windows PCs support local AI agents, image and video generation, and model fine-tuning.
- They offer up to 128GB of unified memory.
- The systems include NVIDIA’s AI software stack and are positioned for RTX-accelerated creative work and development.
What happened
NVIDIA announced on November 18, 2026, that RTX Spark small desktops had launched from ASUS, Dell and MSI. The dedicated Windows PCs support local AI agents, image and video generation, and model fine-tuning. They offer up to 128GB of unified memory.
Why it matters
NVIDIA presented local PCs as platforms not only for running AI agents but also for image and video generation and model fine-tuning. The specification of up to 128GB of unified memory is intended for environments that handle large local models, but the models supported and processing performance will need to be confirmed for each configuration.
Analysis
The product is positioned as a dedicated Windows desktop for running local AI, handling not only inference but also image and video generation and model fine-tuning. Unified memory is specified at up to 128GB, which NVIDIA says can accommodate larger local models. However, this is a maximum, and the announcement does not indicate that every model from each manufacturer will have the same capacity. The information provided is also insufficient to determine model support or processing speed.
The systems will be offered as desktops from ASUS, Dell and MSI, combining NVIDIA’s AI software stack with RTX acceleration. The focus of competition may therefore shift beyond chip performance alone to how each manufacturer combines memory configurations and software environments, and how accessible it makes local generation and fine-tuning. Meanwhile, the published information does not include pricing, specifications for individual models, or power and performance comparisons, so differences among the products cannot be assessed.
To consider the connection with physical AI, it is important to distinguish the design of imitation-learning data covered by APTO selected for student physical AI challenge in GENIAC-PRIZE 2026 from the model fine-tuning promoted for these PCs. The former concerns data design for robot-arm tasks; the latter is an announcement about local computing environments. No use case or collaboration linking the two has been disclosed. If the PCs are applied to robot training, points to confirm include data formats, compatibility with training software, and the process for transferring trained models to physical hardware.
Further assessment will require manufacturer-specific configurations and pricing, supported models and software, performance during fine-tuning, and the operating conditions for local processing. The announcement also did not provide specific GPU configurations, benchmarks or conditions for offline operation. The maximum 128GB capacity alone does not establish what size of model can be handled or at what speed; the actual specifications and operating conditions will determine the products’ uses.
Details
According to NVIDIA, each system includes the company’s AI software stack and memory capacity for handling larger local models. The systems are also positioned for RTX-accelerated creative work and development.
Sources
Translated from the Japanese original: /news/nvidia-rtx-spark-kogata-desukutoppu-wo-hatsubai-asus-dell-ms-5f4a26. The Japanese version is authoritative if they differ.