At a November 6, 2026, event in Seoul, NVIDIA demonstrated RTX Spark laptops handling local AI tasks, including video editing. The reported editing workflow required users to download a suitable model and connect an AI agent to the editing software before running it, according to ChosunBiz.
NeoTeo previously covered Lenovo’s Yoga 9n RTX Spark laptops in an earlier report on the lineup. The Seoul event offered a closer look at how NVIDIA’s local-AI pitch was meant to work in practice.
What NVIDIA demonstrated in Seoul
ChosunBiz reported that one demonstration searched a three-hour video for scenes containing a chosen word. Scene selection took about a minute, and the footage was cut into a three-minute edit. In another editing workflow, the Hermes agent controlled DaVinci Resolve Studio, changed aspect ratios for YouTube and social media, and saved five video files. Mistral models were among those used locally.
The event also included email and calendar organization, plus motion design for a virtual robot in a soccer-field-style simulation. The robot was shown in a virtual environment, not operating in the physical world.
NVIDIA representatives told ChosunBiz that the editing tasks ran on the laptop without an internet connection or paid calls to an external AI service. That offline workflow came after setup, rather than straight out of the box.
The setup behind the offline workflow
The reported sequence was straightforward, but it involved preparation: choose a model suited to the task, download it, then connect the agent to the relevant application. Only after those steps could the demonstrated editing workflow run locally.
For a local AI workflow, count preparation as part of the task: choose and download the model, connect the agent to the app, then run the job.
NVIDIA announced RTX Spark on May 31, 2026, and said laptops and compact desktops were planned for fall 2026. ChosunBiz later reported that NVIDIA officially launched the RTX Spark chip on October 16. Its November 6 Seoul event came after both milestones.
The hardware behind RTX Spark
NVIDIA describes RTX Spark as a Windows PC platform built around a 20-core Grace CPU and a Blackwell RTX GPU with 6,144 CUDA cores. The GPU also has fifth-generation Tensor Cores with FP4 precision. NVIDIA says the platform supports up to 128 GB of unified memory and up to one petaflop of AI compute. Unified memory is a shared pool that the CPU and GPU can use; the maximum does not describe every laptop configuration. NVIDIA’s RTX Spark announcement also identifies NVLink-C2C as the CPU–GPU connection and says MediaTek collaborated on the custom Arm CPU design.
NVIDIA named ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI among the laptop makers for the platform, with Acer and GIGABYTE models to follow. ASUS has published specifications for two specific RTX Spark variants:
| Model | Display | Thickness | Weight | Battery capacity | Maximum SSD storage |
| ASUS ProArt P16 H7607 | 16-inch, 16:10 OLED touchscreen, up to 4K | About 0.51 in | About 3.90 lb | 99.9 Wh | Up to 2 TB |
| ASUS ProArt P14 H7407 | 14-inch, 16:10 OLED touchscreen, up to 3K | About 0.55 in | About 3.26 lb | 90 Wh | Up to 1 TB |
ASUS lists up to 128 GB of unified memory for both models. It also specifies 120 Hz with variable refresh rate, up to 1,600 nits HDR peak brightness and an anti-reflection coating for each display. The P14 is the lighter, smaller-screen option; the P16 has a larger display and a higher maximum SSD capacity. These are ASUS specifications for the named models.
What reviewer testing adds
Linus Tech Tips tested RTX Spark laptops in gaming, productivity, creative software, AI workloads, battery and thermal tests. Its review reported mixed results in gaming and productivity, alongside strong prompt-processing performance in some AI tasks. Reviewer Linus Sebastian concluded that the tested devices were difficult to recommend overall. Those results add a different perspective from the Seoul demonstrations, which showed selected workflows rather than a range of benchmark workloads.
Where local processing may fit
NVIDIA representatives described repeated video-production and robot-training tasks as potential uses for local processing. A senior managing director quoted by ChosunBiz said lighter tasks such as email management could instead use cloud services. NVIDIA’s Jeff Yen said that whether local hardware pays for itself depends on the work being done.
