Gigabyte's AI TOP AERO puts a local AI workstation in a roughly 1-liter desktop

What Gigabyte announced

Gigabyte has announced a desktop computer called AI TOP AERO, built on Nvidia's RTX Spark platform. As technews.tw reported, in a release it carried from PR Newswire, the company pitches it as a "Personal AI Studio": one small Windows machine for developing AI, making content and playing games.

The core idea is to run larger AI models on the device itself instead of renting time on a cloud service. The release says privacy protection is the default, since data and inference (the step where a model produces an answer) stay on the machine.

This is a vendor announcement, so the claims about performance and convenience come from Gigabyte and Nvidia's own framing. Independent testing has not been published in what we have seen.

What is inside the box

The machine is built around the RTX Spark N1X Superchip. It pairs a 20-core Nvidia Grace CPU with a GPU based on Nvidia's Blackwell architecture, the company's current graphics design.

The most important number for AI work is memory. The release lists up to 128GB of LPDDR5x unified memory. "Unified" means the processor and graphics chip share one memory pool instead of each having its own. Local AI models are large, so a big shared pool is what lets bigger ones fit.

Storage goes up to 4TB on a PCIe Gen5 SSD, a fast type of solid-state drive. Ports include USB Type-C and 10GbE wired networking (10 gigabit Ethernet), plus Wi-Fi 7.

Gigabyte says its cooling design fits into a chassis of about 1 liter. AI work often keeps a machine under heavy load for long stretches, and the company says the design leaves enough thermal headroom to hold performance over time. That is a manufacturer claim we cannot check.

The software side

The release leans on Nvidia's software ecosystem. Developers can use CUDA, Nvidia's programming platform, to build, prototype, fine-tune and run open models locally. Nvidia's Nemotron models are named as supported.

It also lists LM Studio and Ollama, two apps for running models on your own computer. For image and video generation it names ComfyUI, a tool for building generative AI workflows.

Gigabyte also pitches the machine for AI agents, programs that carry out tasks with some autonomy. The idea is that an agent can stay available around the clock on your desk, with data kept on the device.

The release adds Nvidia PAIR (Personal AI Router). With it, the desktop can join a local network and share inference jobs with other available devices, which would let a home or studio spread AI work across several machines.

Creators and gamers get a mention too

The machine is not marketed only to developers. Nvidia Broadcast and RTX Video technologies are listed as ways to improve audio and video quality, and the release says the machine can also handle games.

For games, it names DLSS 5, which uses neural rendering, meaning AI techniques that generate more detailed lighting and textures. This works in supported games only. The pitch is that one desktop can switch between AI work, content creation and play.

Who should care, and what is missing

The clearest audience is people who handle confidential material and want to use AI without sending it elsewhere. Think small firms with client files, or developers testing models. A cloud model remains simpler for light, occasional use, and it needs no hardware purchase.

The key gap is cost. The release says configurations, pricing and supply will follow Gigabyte's local announcements in each region. It does not say which memory and storage options come at which price, or how large a model each configuration can run in practice.

Until independent reviews appear, treat the specifications as a menu of what is possible, not proof of how it performs. The launch date, November 10, 2026, leaves time for that evidence to arrive.