Summary
Creator Alex Finn picks up a second NVIDIA DGX Spark from Micro Center and links it to his first to see how far a home AI lab can go. Each Spark carries 128 GB of unified memory, so the pair shares 256 GB, enough to host much larger models than a single unit can.
The video walks through unboxing, the headless Wi-Fi setup that needs no monitor, connecting the machines with a direct link cable, and putting every device on a Tailscale private network. Finn then uses an AI agent to configure GPU settings and download a roughly 400 GB copy of DeepSeek V4.1 Flash, a fast open-weights model that has drawn attention for its speed on dual-Spark setups.
Along the way, he describes his wider home lab, which includes Mac Studios, Mac Minis and an AMD Halo machine. He also shows a practical tip: ask an AI agent to inspect your hardware and recommend which local model fits it best. The central question is whether a two-Spark cluster can run a coding workflow well enough to replace cloud subscriptions such as Claude and OpenAI. This is useful viewing for anyone weighing local AI hardware against paying for cloud models.
📺 Source: Alex Finn · Published October 27, 2026
🏷️ Format: Hands On Build
