GAMES Aspyr releases unannounced Tomb Raider 1-3 Remastered patch 1.14 to address save corruption and animation bugs NINTENDO SWITCH 2 Final Fantasy VII Revelation gives Zack Fair his own combat system and skill tree AMD Ace Combat 8 PC Performance: RTX 5090 Drops to 47FPS in Co-Pilot Mission GAMES MediEvil Recomp Adds Widescreen and Frame Interpolation While Rejecting AI NINTENDO SWITCH 2 Tomb Raider: Legacy of Atlantis Drops Denuvo DRM Ahead of 2027 Launch GAMES Snake Voice Actor Akio Otsuka Urges Konami to Make New Metal Gear Solid Without Kojima NINTENDO Ubisoft files trademark for ‘Motherbrain’, an AI toolset for game development NINTENDO Angelian Trigger arrives on PC December 10, PS5 version delayed until 2027 XBOX SERIES X Hell Let Loose: Vietnam update 1.5 fails to fix console graphics, players compare visuals to PS1 CONSOLES Used PS5 Pro prices hit $1,499 at GameStop as GTA 6 scarcity drives secondary market GAMES The Witcher 4 voice actress Denise Gough recorded dialogue for Ciri, raising Yennefer rumors PHONES Google Pixel 10a price jumps to $599, erasing value gap with Samsung Galaxy A57 ANDROID Vampire Crawlers Mobile Launches October 12 with Full Death Parade Update GAMES Overwatch Season 5 Adds Solo Levelling Anime Skins

Xiaomi AI Cube Prototype Runs 120B LLMs Locally With Custom XRING Silicon

Lena Fischer 3 min read

Xiaomi unveils the AI Cube prototype, a mini- PC with custom XRING silicon designed to run large language models up to 120B parameters locally.

Xiaomi AI Cube prototype device
PC HARDWARE

Xiaomi has introduced the AI Cube, a prototype mini-PC designed to run large language models directly on local hardware. This development matters because it demonstrates a path toward high-capacity AI inference without relying on cloud servers. Users interested in private data handling or offline AI workflows can now see a concrete example of what such a device looks like in practice.

Xiaomi AI Cube prototype device
The Xiaomi AI Cube prototype features an aerospace-grade aluminum unibody.

Prototype mini-PC uses three proprietary chips for local LLM deployment

The device serves as a technical demonstration of Xiaomi's internal silicon strategy. It houses three proprietary chips developed by the company. The main processor is the XRING O3, which acts as the central AI system on a chip. Two other chips handle acceleration and specialized driving tasks respectively.

Specifications

  • AI SoC: XRING O3 (10-core CPU, 16-core G2-Ultra NX GPU, 200 TOPS NPU)
  • AI Acceleration Chip: XRING O100 (6nm 3D wafer-level stacking, 1.22TB/s near-memory computing bandwidth)
  • Intelligent Driving Chip: XRING D100 (3nm, 20-core CPU, 16-core NPU, up to 160GB memory support)
  • Form Factor: Mini-PC prototype, aerospace-grade aluminum unibody, 150W continuous power
  • Model Support: Local deployment of LLMs from 3B to 120B parameters

The XRING O3 features a 10-core all-big-core CPU layout and a 16-core G2-Ultra NX GPU. It also includes a 200 TOPS low-power NPU for efficient neural network processing. The XRING O100 acceleration chip uses 6nm 3D wafer-level stacking to achieve 1.22TB/s near-memory computing bandwidth. The XRING D100 driving chip supports up to 160GB of memory with a 20-core CPU and 16-core NPU.

The hardware enclosure is an aerospace-grade aluminum unibody with 33,874 CNC precision cutouts. It can sustain up to 150W of continuous power output. The system supports the local deployment of large language models ranging from 3B to 120B parameters. This range allows it to handle both lightweight and massive AI workloads on a single machine.

Xiaomi has not announced a commercial release date or pricing for the AI Cube. The company stated that the device is currently a prototype. It remains unclear whether this hardware will become a consumer product or stay as a technical demonstration. We have been tracking Xiaomi AI Cube closely — see our earlier coverage on Xiaomi Mix Fold Next Gen to.

The AI Cube confirms Xiaomi's capability to build custom silicon for high-performance AI tasks. The prototype showcases specific architectural choices like 3D stacking and high-bandwidth memory interfaces. Buyers should watch for official announcements if Xiaomi decides to commercialize this platform.

Source: Xiaomi, GIZMOCHINA