South Korea’s DeepX to spend over $100 million on a 5-watt AI chip built on Samsung’s 2nm process

South Korean AI chip startup DeepX plans to spend more than $100 million developing its next-generation AI chip, the DX-M2, built on Samsung Electronics’ 2-nanometre manufacturing process. The company said the DX-M2 design is complete and has been handed to Samsung’s foundry unit, with first silicon expected in March 2027, customer sampling in the second quarter of 2027, small-volume shipments in the fourth quarter, and mass production in 2028. Samsung declined to comment.

Five watts for large models, aimed at robots and AI PCs

The DX-M2 is a neural processing unit (NPU) aimed at robots, industrial equipment and AI-enabled personal computers, designed to run large language models and vision-language models on about 5 watts — a design target, not measured performance. The company disclosed no performance figures, dimensions, prices or customer names. DeepX also signed a memorandum of understanding with Intel Korea on the same day to bring its NPUs into Intel’s OpenVINO software, and is working with Intel Korea and South Korean server maker KTNF on local AI inference server business.

The previous generation is already in production, with 88 orders

DeepX’s previous chip, the DX-M1, is built on a 5nm process and entered production in August 2025; as of October 2026 it had won 88 orders worth about $13.3 million across 10 countries. The DX-M2 is the next chip in the same line and DeepX’s first to use Samsung’s most advanced second-generation 2nm node (SF2P).

(Cover image is a file photo of a Samsung chip fabrication cleanroom.)