Last updated: November 10th, 2026 at 08:58 UTC+01:00


New AI chip made using Samsung's 2nm tech uses just 5 watts of power

Samsung's chip manufacturing quality seems to be improving.

Asif Iqbal Shaik

Reading time: 1 minutes

DEEPX DX-M2 On-Device AI Chip

DEEPX

Business

DEEPX DX-M2 on-device AI inference chip

TL;DR

  • DEEPX unveiled the DX-M2, an on-device AI inference chip built on Samsung Foundry’s 2nm GAA process.
  • The chip is designed for industrial devices and robotics, and DEEPX says it uses about 5 watts of power.
  • The DX-M2 supports on-device LLM and VLM inference, with tape-out complete and mass production planned for 2027 and 2028.

After facing several setbacks in contract chip manufacturing business over the past few years, Samsung's foundry division is getting back on its feet. Beyond improving its technology, the company is winning new clients. In the latest win, a new on-device AI inference chip has been announced that is fabricated by Samsung Foundry and consumes just 5W of power.

DEEPX's 2nm AI chip made by Samsung uses just 5W of power

DEEPX, a South Korean fabless AI chip firm, has unveiled its next-generation AI chip, the DX-M2. Designed for industrial devices and robotics, the chip is built on Samsung Foundry's cutting-edge 2nm GAA process node. DEEPX claims the chip consumes approximately 5W, and thanks to that efficiency, it has earned a CES 2027 Innovation Award in the AI category.

The DX-M2 is engineered to run on-device inference for Large Language Models (LLMs) and Vision Language Models (VLMs), enabling industrial machines and robots to operate autonomously without relying on the cloud or external data centers.

The chip completed tape-out earlier this month, with initial mass production slated for the fourth quarter (Q4) of 2027, followed by full-scale mass production in 2028.