According to reports from South Korean media, South Korea's fabless AI chip company DEEPX has officially launched its next-generation edge AI inference chip, the DX-M2. This chip is designed for industrial equipment and robots, manufactured using Samsung's 2nm GAA process. DEEPX states that the power consumption of the DX-M2 is approximately 5 watts. Due to its excellent energy efficiency, the chip has been awarded the CES 2027 Innovation Award in the AI category. For industrial applications, power consumption directly affects whether equipment can operate offline for extended periods; 5 watts not only represents energy savings but also implies that heat dissipation and power supply designs can be further lightweighted.
In terms of functionality, the DX-M2 focuses on edge inference, capable of running large language models and visual language models, allowing industrial machinery and robots to operate autonomously without relying on cloud or external data centers. This approach aligns with recent trends in edge AI development: placing computational power locally within devices to reduce latency and reliance on networks while ensuring that data does not need to leave the site.
DEEPX DX-M2 Chip Expected to Achieve Full Production in 2028
Regarding the timeline, DEEPX has stated that the chip completed its tape-out earlier this month, with initial mass production scheduled for the fourth quarter of 2027, followed by full-scale production in 2028. There is approximately a two-year window from tape-out to large-scale shipment, during which process validation needs to be completed, and the industrial customers' system designs must catch up; thus, the actual implementation timeline still depends on downstream progress.

