First Home Server Build Targets AI Hosting

This is a solid “learn by shipping” box, not a serious training rig yet. If the real near-term goal is MVPs plus API-driven agent workflows, the smartest spend is reliability and headroom, not rushing into a GPU purchase.

• –`i5-13400` is fine for containers, databases, and web apps, but `32GB RAM` will get tight once you stack observability, background jobs, vector stores, and model-serving sidecars
• –`RTX 4060 Ti 16GB` is acceptable for 7B-13B quantized inference, but it’s bandwidth-limited and not the card you buy for meaningful SLM training
• –`1TB SSD` is the first spec I’d upgrade; model files, Docker layers, logs, and datasets eat space embarrassingly fast
• –`750W PSU` is more than enough, which is good for future GPU swaps, but 24/7 efficiency and noise matter more than raw wattage
• –If local AI is still “maybe later,” skip the GPU for now and put the savings into `64GB RAM`, a larger NVMe drive, and better cooling