Mobilint is a Korea-based AI semiconductor company building power-efficient NPUs for on-device and on-premise inference, aiming to shift workloads from cloud GPUs into compact systems with predictable latency and power. In the interview they mention working across the memory and foundry ecosystem (including Samsung and SK hynix) while focusing the demo on ARIES: 80 TOPS in a 25 W TDP, PCI Express Gen4 x8, and 16 GB LPDDR4X (optional 32 GB) with 66.7 GB/s memory bandwidth https://www.mobilint.com/
On the demo table, ARIES is framed through deployable computer-vision throughput: YOLO-11 object detection plus standard backbones like ResNet-50 and MobileNet, with attention on TOPS per watt rather than peak TOPS alone. The target is industrial PCs and compact edge servers where thermal headroom is tight, so inference stays local while multiple models share one host chip.
A second setup zooms out to a larger PCIe card concept that “crams” four Mobilint M800 accelerators onto one board, intended to run several ~8B-parameter language models concurrently, or scale up via partitioning and batching. That naturally leads to vision-language models: camera frames become embeddings, the text decoder turns them into scene descriptions, and multilingual output becomes a useful interface for inspection or support, recorded on the CES Las Vegas 2026 show floor there.
For smaller, always-on endpoints, Mobilint highlights REGULUS, a full SoC that pairs an NPU with Arm Cortex-A53 CPU cores so it can run Linux and execute pre-trained models without a separate host. They cite around 10 TOPS under 3 W for drones, robots, and AI CCTV, then demonstrate high-input video analytics, including a 96-stream fire-risk example where bandwidth, buffering, and scheduling matter as much as raw compute in the field.
The closing theme is vehicle and humanoid readiness: partners want edge AI that is fast and power-bounded, but also engineered for functional safety and security hardening, not just benchmarks. The takeaway is that autonomy progress is a mix of smarter models, tighter sensor-to-actuation pipelines, and consolidating silicon so the platform can scale compute without multiplying energy cost today.
I’m publishing about 100+ videos from CES 2026, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Check out all my CES 2026 videos in my playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjaMwKMgLb6ja_yZuano19e
This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK
Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁
Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY
