I stopped renting GPUs: two boxes and an iPhone app run my whole AI stack

EAccelerate_42 · x · 2026-09-05

The author ditched rented GPUs for two small boxes in a closet (121 GiB RAM + 48 GB VRAM) controlled via an iPhone app. Local stack includes DeepSeek V4 Flash (384K context), Qwen3.8 27B, GLM-5.3 Flash, Qwen-Image with editing, and MiniMax H3 video. Access is Tailscale-only, with zero tokens billed.

Original post →

More from Infra

Infra channel →