My €1,000 used-hardware rack runs my home services, production apps and local inference

My rack cost €1,000 in total, all used hardware from eBay Germany. 2017 repurposed office PC running proxmox and a used Facebook marketplace build hosting an AMD RX 6900 XT with 16GB of VRAM bare metal, no virtualization. It runs my home network services and supports production apps used on iOS, Android and the web. Proxmox handles VMs and LXC containers, with Docker for application services. The setup includes OPNsense, Pi-hole, shared storage, backups, Immich, a mail archive, and a small Kubernetes cluster with Gitea and a container registry. The GPU runs on a separate bare-metal machine. Text inference uses llama.cpp. The applications cover mobile assistants, book production, image and wallpaper generation, music generation and mastering, and an AI news service, all with the assistance of a diffuser forma image and ACE for audio. Most of the app backends live on the Proxmox side and call the inference services over the LAN. The interesting part has been sharing the resources. Model weights fitting in VRAM is only the start: context, concurrent requests and image generation also need room. Some workloads coexist; heavier generation jobs take turns. Selected services have CPU fallbacks when the GPU endpoint is unavailable. Public apps use Cloudflare tunnels. Monitoring and remote access are part of the setup too, with a separate Raspberry Pi connected to the hypervisor’s serial console for recovery when normal management access fails. I built much of the application layer with AI-assisted coding. Keeping it useful for actual users has meant working on queues, workload profiles, health checks and fallbacks as much as the apps themselves. Are you guys still buying big, chunky enterprise servers for self-hosting, or do you prefer used office PCs with a five- or six-year-old consumer GPU for inference? For those who chose enterprise hardware, what made it worth it for your services: RAM capacity, drive bays, PCIe expansion, remote management, or something else?