Broken phones emerge as cheap home inference nodes

Using discarded flagship phones as dedicated inference nodes is a brilliantly scrappy hack that highlights the untapped compute power sitting in drawers.

• –Flagship mobile chips like the Snapdragon 8 Gen 2 punch above their weight for small model inference.
• –Sourcing provider-locked or screen-damaged devices drastically undercuts the cost of dedicated server GPUs.
• –Thermals and battery management remain the biggest hurdles for running a 24/7 mobile inference cluster.
• –This trend highlights the growing viability of edge AI hardware for self-hosted LLM setups.