Tags: 0 - HomeLab 0 - Projects Networking-Computer
2026 Homelab Overview
My 2026 homelab is a layered infrastructure environment designed to support local AI inference, media services, storage, and hands-on platform engineering work.
The core design decision is separation of concerns. Compute, control, data, and user access each have a defined place in the system so that experimentation does not compromise stability and production-like services do not interfere with learning work. It is also the environment I am using to move toward MLOps engineering by learning how AI serving infrastructure behaves in a system that has to be designed and operated deliberately.
Core Layers
- AI compute layer for local GPU inference
- Control layer for routing, reverse proxying, and orchestration
- Data layer for media, storage, and downloads
- Interface layer for workstation access and development
What This Environment Supports
- Local-first AI inference with cloud fallback
- Stable media services using the ARR stack
- Safe RHCSA and Rocky Linux experimentation
- A practical base for future platform engineering projects
- A hands-on path into MLOps through model serving, routing, and AI infrastructure operations
Current Node Roles
AI Node
Dedicated GPU compute host for model serving and inference workloads.
- Platform: AM4
- GPU: RTX 4070 Ti Super with 16 GB VRAM
- RAM target: 64 GB
- OS: Ubuntu Server 24.04
- Primary services:
vLLM,Ollama - Future service:
ComfyUI
Control and Lab Node
The control-plane host runs the gateway, virtualization layer, and practice environments used for operating system and infrastructure work.
- Hardware: Optiplex 3080 SFF
- CPU: Intel i5
- RAM target: 32 to 64 GB
- OS: Proxmox
Media and Storage Node
A stable data platform for media serving, download automation, and photo management.
- Hardware: HP Elitedesk G4
- RAM target: 16 to 32 GB
- Storage: 2 x 4 TB HDD plus 256 GB NVMe
- OS: Ubuntu Server 24.04
- Primary services: Jellyfin, Sonarr, Radarr, Prowlarr, qBittorrent, Immich
Main Workstation
The workstation is the primary operator interface into the environment rather than part of the server layer itself.
- Platform: AM5
- GPU: RX 7800 XT or equivalent
- OS: Windows with optional dual boot
- Usage: SSH, VS Code, Parsec, gaming
Design Principles
One Machine, One Job
- The AI node handles compute
- The Optiplex handles control and experimentation
- The Elitedesk handles data and media
Local-First AI
Local GPU inference is the default path. Cloud APIs are used as fallback rather than as the primary architecture.
Stable Systems Stay Stable
Stable services stay on stable nodes. Experimental work belongs on the lab infrastructure.
Minimal Complexity First
Additional orchestration only gets introduced when it solves a real operational problem. Complexity is treated as a cost, not a milestone.