155 points24 comments
DeepX’s DX-M2 runs 100B-parameter LLMs on edge under 5 W
What is DeepX claims inference up to 100B parameters at less than 5 W?
The DX-M2 is a second-generation AI accelerator that uses Samsung's processor-in-memory technology to run language models of up to 100 billion parameters while consuming less than 5 watts. It is aimed at robots, drones and other low-power edge devices, delivering roughly 10-30 tokens per second.
Who it is for
Robotics and edge-AI engineers can now consider on-device generative AI without needing a power-hungry server.
The story so far
Early Signal first spotted DeepX claims inference up to 100B parameters at less than 5 W on November 10, 2026, with 155 points. Early Signal files it under Inference.