Home > Business > Industry

DeepX to produce next AI chip using Samsung Foundry’s 2-nanometer process

Published: 14 Apr. 2026, 17:34 Updated: 14 Apr. 2026, 19:04

LEE JAE-LIM
[email protected]

DeepX CEO Kim Lok-won speaks at the company's press conference at its Pangyo office in Gyeonggi on April 14. [DEEPX]

DeepX, a Korean chip design startup valued at more than 1 trillion won ($677 million), will start production of its next AI chip next year using Samsung Foundry’s 2-nanometer process.

The chip is known as DX-M2, DeepX’s second low-power AI processor for on-device AI and large language model inference. It targets delivery of up to 80 trillion operations per second while consuming less than 5 watts of power.

“I believe the physical AI market will become even larger than the data center market,” Kim said on Tuesday at a press conference at the company’s Pangyo headquarters in Gyeonggi, where DeepX unveiled its product road map. “We see ourselves not simply as an AI chip design company, but as a physical AI infrastructure company powering everything from small toy robots to cars, autonomous vehicles and high-performance humanoids.”

Alongside Rebellions and FuriosaAI, both of which focus on data center AI processors, DeepX is widely regarded as one of Korea’s leading AI chip unicorns — privately held startups valued at more than 1 trillion won.

DeepX is betting that physical AI will be the next major growth market and is developing ultra-low-power chips capable of running advanced AI models directly inside robots, smart mobility platforms, intelligent cameras and other connected devices that have traditionally relied on cloud-based data center infrastructure. The company argues that growing power constraints will accelerate the shift toward on-device AI.

Its first product, DX-M1, which entered mass production in August 2025, consumes an average of 2 to 3 watts and delivers 20 times greater power efficiency than Nvidia’s on-device GPU, Jetson Orin, while priced at roughly one-tenth the cost, according to the company. DeepX also said it achieved a yield over 90 percent on Samsung’s 5-nanometer process for the chip.

The company said it has secured more than 30 production orders within seven months of starting mass production of DX-M1, spanning eight industries, including robotics, smart factories and mobility. Cumulative order value currently stands at about $6.6 million.

Among its customers is Hyundai Motor Group. DeepX said an AI computing solution co-developed with Hyundai Motor Group’s Robotics Lab has completed production validation and will be deployed in the delivery robot DAL-e and mobility platform MobED, with mass production scheduled to begin later this year.

The company is also working with Baidu to jointly develop application algorithms. DeepX chips are being integrated into Baidu’s optical character recognition (OCR) cameras, data parser systems and drones, according to the company.

DeepX is further developing software compatible with Nvidia’s robotics development platform, Isaac. The aim is to allow customers to maintain their existing software environment while replacing the underlying GPU hardware with more affordable, power-efficient DeepX chips.

For 2026, the company is targeting total revenue of $40 million, including more than $25 million from product sales. It also aims to achieve annual profitability on the back of DX-M1 sales.

DeepX said it is also preparing for a domestic listing and plans to appoint lead underwriters after completing its current fund-raising round.

BY LEE JAE-LIM [[email protected]]