The Ultimate Hypothetical AI PC, Part By Part
Build the ultimate local AI PC: 4 RTX PRO 6000 Max-Q cards, 384 GB VRAM, 2 TB RAM, 96-core Threadripper, $136k parts breakdown with real speeds. The ultimate personal AI computer for running open-weight models at home pairs four NVIDIA RTX PRO 6000 Blackwell Max-Q graphics cards (96 GB GDDR7 each, 384 GB total) with an AMD Threadripper PRO 9995WX CPU, 2 TB DDR5 system RAM, and a 2200-watt power supply, approximately $136,089 in parts listed October 11, 2026. The build fits the fourth-best open model (GLM-5.3 Flash at 204 GB NVFP4) entirely on GPU memory for 63.7 tokens/sec on a single request, but the top three ranked models (MiMo-V2.6-Pro, GLM-5.3 full, Kimi K3) require overflow into system RAM, where DDR5 memory bandwidth sets the pace: Kimi K3's arithmetic ceiling is around 14 tokens/sec, and real-world software lands below that. The Max-Q variant's 300 W blower cooler passes four-card thermal validation on air (versus the 600 W Workstation Edition, which fails); both share identical 1,792 GB/s memory bandwidth, making compute sacrifice negligible for text generation. At full load, the machine draws 1.9 kW at the wall, exceeding standard 15 A outlets and requiring a dedicated 240 V circuit plus an electrician. Alternatives include NVIDIA's DGX Station (748 GB coherent memory, $92,550, holds top two models but not Kimi K3), BIZON's pre-built four-Max-Q system ($147,125 with 1 TB RAM), Mac Studio M5 Ultra (256 GB, $10,799; 512 GB coming late October), or renting four RTX PRO 6000s at $7.20/hour ($136k buys ~18,900 hours). This machine only makes sense if you need the largest open-weight models running locally under your own roof. For builders, system integrators, and anyone evaluating whether to buy, rent, or build for serious local LLM inference. Chapters: 0:00 Intro 1:40 The Cards 4:50 What Fits 8:19 The Memory 11:36 The Platform 13:22 The Power 15:19 Buy Instead 18:27 Verdict Tools & resources mentioned: - NVIDIA RTX PRO 6000 Blackwell Max-Q: https://www.nvidia.com/en-us/products/workstations/professional-desktop-gpus/rtx-pro-6000-max-q/ - AMD Threadripper PRO 9995WX: https://www.amd.com/en/products/processors/workstations/ryzen-threadripper/9000-wx-series/amd-ryzen-threadripper-pro-9995wx.html - ASUS Pro WS WRX90E-SAGE SE: https://www.asus.com/motherboards-components/motherboards/workstation/pro-ws-wrx90e-sage-se/techspec/ - NVIDIA H200 NVL: https://www.nvidia.com/en-us/data-center/h200/ - NVIDIA DGX Station GB300: https://www.nvidia.com/en-us/products/workstations/dgx-station/ - SGLang: https://github.com/sgl-project/sglang - llama.cpp: https://github.com/ggml-org/llama.cpp - Artificial Analysis LLM Rankings: https://artificialanalysis.ai/ About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #pc build #local ai #llm #rtx 5090 #mac studio
