Ryzen AI Max PRO 400 Systems Ship

AMD is making memory capacity the defining advantage of local AI PCs, shifting the conversation from whether models fit to whether organizations can run them privately and economically.

• –Up to 160GB of GPU-addressable memory makes massive local models practical without discrete workstation GPUs.
• –“Fits locally” does not guarantee fast inference; bandwidth, cooling, and sustained power remain important constraints.
• –AMD’s support for ROCm, PyTorch, vLLM, llama.cpp, Ollama, and LM Studio gives developers a credible local experimentation stack.
• –The $7,449 HP ZBook Ultra G3a shows that this is initially premium infrastructure, not an immediate replacement for cloud inference.