DEEPX Announces 2nm DX-M2 Edge AI Chip with 80 TOPS NPU
Korean edge AI chip developer DEEPX announced its next-generation DX-M2 chip, which is manufactured using Samsung's 2nm process and delivers 80 TOPS of NPU performance at a 5W power consumption. The physical chips are expected to be available in early 2027 following a recent tape-out milestone. This development marks a significant step forward in bringing advanced node technology and power-efficient large language model capabilities to edge devices. It enables local execution of models with over 10 billion parameters within a strict 5W power budget. The DX-M2 supports up to 128-bit LPDDR5X-9600 memory and will be available in single-chip variants as well as versions with packaged integrated memory. It has also received an innovation award in the artificial intelligence category for CES 2027.
## BACKGROUND
Edge AI chips like NPUs (Neural Processing Units) are specialized processors designed to run machine learning tasks locally on devices rather than relying on cloud servers, reducing latency and bandwidth usage. DEEPX previously developed products like the DX-M using a 5nm process, while advanced manufacturing processes like Samsung's 2nm allow for higher transistor density and better energy efficiency.