4 Mac Minis run a 70B model that no single one could hold.
Gold aluminum boxes standing upright on a $180 rack shelf. Fans humming underneath. Cables tangled across the floor like roots.
He bought each Mini used. $429 apiece. Base M4, 16GB RAM. Alone, nothing special.
Wired together over Thunderbolt, running MLX in cluster mode, splitting one 70B model across all 4 nodes. 19 tokens a second. Fully local. No API call leaves the room.
He used to pay $340 a month in OpenAI credits building chatbots for clients. Now the bill is electricity. $11.
Clients never see the rig. They see a private, on-prem AI system — the kind enterprises pay $50,000 consultants to architect.
He charges $1,200 a build. Booked through November.
He's not selling a chatbot. He's selling enterprise infrastructure that fits on a shelf.