CHUWI Unveils 192GB RAM Mini PC for Local 300B LLM Inference

CHUWI unveiled the UniBox AI495 Pro at the 2026 Apsara Conference, offering a compact solution for running large language models locally. This shift matters because it allows users to bypass recurring cloud API subscription fees for generative AI tasks. The company positions the device as a direct alternative to expensive cloud inference services.

Compact 2.9L device offers 192GB memory for local AI inference

The AI495 Pro houses an AMD Ryzen AI Max+ PRO 495 processor with 16 cores and 32 threads. It reaches boost speeds of 5.2GHz and pairs with a massive 192GB of LPDDR5x-8533 unified memory. A lower-end model, the UniBox AI395, uses the Ryzen AI Max+ 395 chip and 128GB of memory.

Specifications

• CPU: AMD Ryzen AI Max+ PRO 495 (16-core, 32-thread, up to 5.2GHz)
• Memory: 192GB LPDDR5x-8533 (273GB/s bandwidth)
• GPU: Radeon 8065S (up to 160GB dedicated VRAM)
• Dimensions: 209.6×209.6×67mm (2.9L volume)
• Storage: 2TB SSD (M.2)

CHUWI allocates up to 160GB of that unified memory as dedicated VRAM for the integrated Radeon 8065S GPU. The system supports two M.2 SSD slots and ships with a 2TB drive. Connectivity includes dual 10GbE ports, dual USB4 interfaces, OCuLink, and Wi-Fi 7.

The device measures 209.6 by 209.6 by 67mm, fitting into a 2.9-liter X-shaped metal chassis. CHUWI claims the AI495 Pro can run 300B+ parameter models like Deepseek V4 Flash 284B in 4-bit quantization. We have tracked the UniBox AI495 Pro closely — see our earlier coverage on CHUWI UniBox AI495 Pro Brings 192GB.

CHUWI stated, "Goodbye Token bills! CHUWI UniBox series AI computing boxes debut at 2026 Apsara Conference." The company also demonstrated the lower-spec AI395 achieving 40 tokens per second during live testing. The vendor confirmed these figures represent local inference capabilities rather than cloud-based processing.

The UniBox AI495 Pro arrives globally in late October 2026. CHUWI plans to announce the final pricing closer to the release date. The device will support Windows 11 Pro as an optional operating system configuration.

This hardware targets developers and enterprises that require high-capacity local AI inference without cloud dependency. The 192GB memory pool enables model sizes that standard consumer PCs cannot accommodate. Users can expect a compact form factor that replaces traditional desktop towers for AI workloads.