User Builds Three 4x R9700 Local AI Servers with 384GB Total VRAM

To power AI for his USA Arms business, a user built three servers each with four Radeon AI PRO R9700 GPUs, costing about $10.7k per machine and totaling 384GB VRAM. After hosted models refused certain research tasks, he moved to local AI and chose Qwen for best cost-performance after testing.

• Open-weight models edge past Sonnet 5 on some benchmarks2026-10-16
• China proposes export controls on 23 AI technologies2026-10-16
• Mapping post-Transformer architecture landscape in 20262026-10-15
• Company Serves 40 People with Local AI on 4x 6000 Pro GPUs2026-10-14
• Someone runs Qwen 27B at ~20 tok/s on a jailbroken PS5 GPU2026-10-14