Local AI over USB: what ASUS UGen300 offers
Adding artificial intelligence capabilities to an existing computer rather than replacing it is the idea behind ASUS UGen300, a USB-C accelerator covered by Sky TG24 on October 30. The device is intended to run compatible models close to the data, with applications including audio transcription and image recognition.[1]
This is not a November 9 announcement: ASUS introduced it on April 1, 2026. The recent Italian coverage nevertheless provides a timely opportunity to examine an alternative to cloud-only services: adding a dedicated processor for inference, the execution of already trained models, alongside an existing computer.[1][2]
According to ASUS, UGen300 combines a Hailo-10H chip with 8 GB of LPDDR4 memory and delivers up to 40 TOPS at INT4, with stated typical power consumption of 2.5 watts. These are manufacturer specifications, not independent test results, and should not be read as guarantees about the speed of every application.[3]
Software is the practical issue. As Sky TG24 explains, the task needs an appropriate model and application; plugging in an accelerator does not make every assistant compatible.[1] Platform support also deserves scrutiny. ASUS lists Windows, Linux and Android, while retaining notes about Windows drivers and Android availability initially aimed at business customers.[2][3]
Local processing can avoid sending inference data to the cloud under the architecture ASUS describes.[3] For anyone considering a purchase, the useful question is therefore specific: which everyday tasks actually work with this combination of hardware, models and applications? That is a more helpful buying criterion than the TOPS figure alone.
## Sources
[1] https://tg24.sky.it/tecnologia/now/2026/10/30/asus-ugen300-novita
[2] https://press.asus.com/news/press-releases/asus-ugen300-usb-ai-accelerator-generative-ai-edge
[3] https://www.asus.com/us/motherboards-components/ai-accelerator/ugen/ugen300-usb-8g