ASUS UGen300: USB AI акселератор для периферийных вычислений

When you think of AI accelerators, you usually picture something large and expensive – a graphics card, a server module, or at least an expansion card. ASUS is offering a different format: a small USB dongle that connects to any device and handles artificial intelligence tasks.

On April 1, 2026, the company announced the UGen300 USB AI Accelerator – a device ASUS is calling the world's first USB accelerator for edge AI (that is, AI running directly on the device, not in the cloud). In addition to the USB version, an M.2 form factor is also available for those who prefer a built-in solution.

Что внутри ASUS UGen300

What's Inside the Little Box

The UGen300 is a device measuring 105 × 50 × 18 mm. In terms of size, it's roughly like a thick external SSD, only a bit smaller. Inside is the Hailo-10H processor, specifically designed for AI tasks. Its performance is rated at 40 AI TOPS. To put it simply, this is a unit that measures how many operations a chip can perform per second when working with neural networks. 40 TOPS is enough to run language models, models that work with images and text simultaneously, and other modern AI tools.

In addition to the processor, the device is equipped with 8 GB of its own LPDDR4 RAM, not borrowed from the host system. This is an important point: the UGen300 doesn't 'eat up' resources from the computer or smartphone it's connected to, but operates autonomously within the scope of its tasks.

It connects via USB-C. Power consumption under a typical load is just 2.5 watts. For comparison, a standard night light bulb uses more energy.

Для кого предназначен USB AI-акселератор

Who Is It For and Why

The main idea behind the device is to allow those without powerful hardware to run AI tasks. An old laptop, a single-board computer, an Android device – just plug in the UGen300, and the system gets hardware acceleration for neural networks.

Moreover, there's no need for manual setup: the device is plug-and-play, meaning you just plug it in and it works. It's compatible with Windows, Linux, and Android. It also supports common AI frameworks – the popular tools developers use to create and run neural networks.

The UGen300 can be used for a wide range of tasks: text generation, video summarization, speech recognition and its conversion into commands, and real-time visual information processing.

Преимущества локального ИИ-акселератора

No Cloud – It's Not Just Marketing

One of the key features of the UGen300 is its complete independence from cloud services. This means several things at once.

First, there are no subscriptions. Most cloud AI services charge for usage – either monthly or per request. The UGen300 operates locally, without any recurring payments.

Second, there's no latency. When a request travels to the cloud and back, it takes time. With local processing, the response is generated instantly on the device.

Third, privacy. Your data isn't sent anywhere – it's processed right on the device. For tasks where confidentiality is crucial, this is a significant advantage.

"By integrating the Hailo-10H into a ubiquitous USB device, ASUS is making the full power of AI and generative AI accessible to everyone. We can't wait to see how the developer community will use this plug-and-play accelerator to push the boundaries of AI on devices. This is how Hailo envisions the future of AI: accessible, affordable, and designed for everyone to work with," said Max Glover, Chief Revenue Officer at Hailo.

Возможности и применение ASUS UGen300

What This Changes – and for Whom

The UGen300 appeals to several groups of people.

For developers, it's a convenient tool for experimenting with local AI models without having to buy expensive hardware or pay for the cloud.

For enthusiasts and smart home hobbyists, it's a way to add AI capabilities to devices not originally designed for them.

For businesses that handle sensitive data, it's a solution for local AI without the risk of data leaks to external services.

However, a number of open questions remain. How smoothly heavy language models run with a limit of 8 GB of memory depends on the specific model and task. A performance of 40 TOPS is a good metric for a device of this class, but it's still no substitute for a full-fledged discrete graphics card. Instead, it's a tool for scenarios where portability, power efficiency, and autonomy are more important than maximum generation speed.

Overall, the UGen300 seems like an attempt to reframe our understanding of where and how AI can operate. Not in a data center, not on a powerful desktop – but in your pocket, at the end of a USB cable, next to any device you already have at hand.