Asus has introduced a compact new way to add AI horsepower to a wide range of devices: a USB-based AI accelerator designed to run machine learning workloads locally without the usual hassle of installing a full-size expansion card. Called the Asus UGen300, the device looks and behaves like a USB stick, but it’s built specifically to speed up AI model inference for tasks like image recognition, pattern detection, and other on-device AI workflows.
Unlike traditional accelerator cards that require a PCIe slot and opening up a desktop PC, the UGen300 plugs in externally and connects over USB 3.1 Gen 2. That means it can be a practical upgrade for existing systems, compact computers, or situations where internal expansion simply isn’t an option. It also makes the UGen300 appealing for developers and businesses that want portable AI acceleration they can move between machines.
At the core of the UGen300 is a Hailo 10H NPU. Asus claims up to 40 TOPS of performance using INT4 precision, positioning it as a serious step up for AI acceleration in a small form factor. Power efficiency is another major selling point: typical power consumption is listed at just 2.5 watts, which is low enough to suit always-on edge AI use cases or power-conscious deployments.
The device includes 8 GB of LPDDR4 memory. While that won’t satisfy advanced users working with very large models that demand massive memory pools, it’s more aligned with practical edge AI tasks and pre-trained model inference—especially where the goal is fast, local results without relying on cloud processing.
Compatibility is broad, according to Asus. The UGen300 is said to work across x86 and ARM platforms and supports Windows, Linux, and Android. It also supports common AI frameworks including TensorFlow, PyTorch, and ONNX, helping it fit into existing development workflows. Asus also notes that more than 100 pre-trained models are available, which could reduce setup time for teams aiming to test or deploy AI features quickly.
Another notable detail is cooling. The unit appears to rely on passive cooling, which can be beneficial for silent operation and for deployments where moving parts are undesirable.
Asus has not yet shared pricing or a release date. Still, the concept is clear: a plug-and-play USB AI accelerator aimed at edge AI and local inference, built for users who want a simple way to add NPU performance to PCs and other compatible devices without rebuilding their hardware setup.
