Press release

DEEPX Wins 77 Global Orders Worth Over $13 Million in First Year of Korean NPU Mass Production

AI summary

In the year since mass-producing DX-M1, DEEPX secured 77 purchase orders worth over $13 million across more than 10 countries and regions.

Key points
  • Commercial purchase orders from more than 10 countries and regions in the year since DX-M1 mass production began
  • Business in eight physical and edge AI industries, with more than half of total revenue coming from exports
DEEPX DX-M1, DX-M1M and DX-M1ML chips with block diagrams on an exhibit panel
DEEPX's DX-M1 family of on-device AI chips. (DEEPX)

Aug. 14, 2026 — DEEPX, a developer of ultra-low-power AI semiconductors led by CEO Lokwon Kim, said Aug. 14 that in the year since it began mass-producing its DX-M1 on-device AI chip, it has secured 77 commercial purchase orders (POs) worth more than $13 million from more than 10 countries and regions.

Since starting mass production of DX-M1, an ultra-low-power NPU for on-device AI, in August 2025, DEEPX has won orders in major markets including South Korea, the U.S., China, Europe, Taiwan, Hong Kong, Japan, Singapore and India. Its business spans eight physical and edge AI industries: robotics; smart factories and manufacturing automation; smart mobility and ITS; mission-critical systems and aviation; AIoT; medical; AI IT services and edge computing; and video surveillance and security.

As of the end of July, more than half of the company's cumulative revenue came from overseas, and as global orders turn into actual exports and sales, DEEPX is establishing itself as an exporter of Korean-made NPUs.

62% of All Orders Came in the Last Four Months

DX-M1 is the first on-device AI chip DEEPX has put into mass production. Cumulative POs stood at two in the third quarter of 2025, when mass production began, and rose to six in the fourth quarter, 29 by the end of March 2026, 56 by the end of May and 77 by the end of July.

Notably, 48 of the 77 orders came in the four months from April to July 2026. That is about 62% of all orders, showing that customer evaluations and product designs since the start of mass production have been turning into purchase decisions faster in recent months.

The growth in orders was not concentrated in any single country or industry. First orders from new overseas customers kept rising, and some early customers placed follow-on orders after receiving their products. DEEPX said this shows its AI chip business is moving beyond technology validation and proofs of concept (PoCs) into customer product development, production and supply.

Products supplied include the DX-M1 family — DX-M1, DX-M1 M.2, DX-M1M and DX-M1M M.2 — and some DX-H1 products, delivered as chips or modules depending on customer needs. DX-M1 is an NPU for on-device AI designed to run AI computing on the device itself without a cloud connection, and it is mass-produced on Samsung Electronics' 5nm process. It is aimed mainly at edge AI systems that need to run video and vision AI on low power, such as CCTV, smart factories and robotics.

Robotics, Drones, Smart Factories, Aviation, Medical: Business Expands to Eight Industries

Products using DX-M1 include drones, machine vision cameras and AI boxes, robots and industrial equipment, AIoT devices, medical AI devices, OCR servers for document recognition, video management system (VMS) servers for video monitoring and edge boxes for security cameras. Beyond supplying development kits and evaluation products, some cases are leading to customers' actual product development, production and, in some instances, deliveries.

DEEPX's global customer base is also growing quickly. In China, the company is involved in document recognition (OCR) and machine vision projects related to Baidu, and in AI edge server projects related to global IT and server companies such as Lenovo. In Taiwan, it is expanding to local customers including industrial computing company AAEON. In North America and Europe, supply continues through global distributors and local partners, and in Japan it is expanding new projects, mainly in manufacturing and industry.

400 Companies Evaluated, 27 Global Distributors: Groundwork Laid Before Mass Production

Before mass production, DEEPX spent two to three years running sample-based pre-evaluations with more than 400 companies, building product evaluation experience and a base of potential customers.

After mass production began, DEEPX signed contracts with 27 global semiconductor distribution partners to build an overseas supply chain and worked with about 50 global hardware and module companies to broaden its lineup of application modules. It also linked up with open platform ecosystems such as Raspberry Pi, Baidu's PaddlePaddle and Ultralytics to lower the barrier for customers to adopt DEEPX NPUs.

More Than 500 Patents and 90% Production Yield: From Technology to Exports

DEEPX holds more than 500 patents in NPU design and low-power AI computing, and achieved a yield of about 90% in DX-M1 mass production, securing supply stability and cost competitiveness.

"What we find most meaningful about our first year of mass production is that real orders and repeat orders are coming from many markets around the world, not just one country or industry. An AI chip only proves its competitiveness when it goes into customers' actual products and leads to ongoing orders." — Lokwon Kim, CEO, DEEPX

"Building on the mass-production experience and global customer base we have gained over the past year, we will expand AI chip technology developed in Korea into the global market in earnest." — Lokwon Kim, CEO, DEEPX

Targeting the Next-Generation Physical AI Market With Samsung 2nm-Based DX-M2

DEEPX is also developing DX-M2, a next-generation AI chip built on Samsung Electronics' 2nm process. Under its current plan, the company will receive prototype wafers in March 2027, go through bring-up to power on the chip and verify performance in April and May, and begin preparing to supply early customers in June.

DX-M2 aims to run generative AI on up to about 5 watts at the chip level. It targets power-constrained physical AI products such as robots, drones, intelligent industrial equipment and battery-powered smart devices.

  • #deepx
  • #npu
  • #dxm1
  • #ondeviceai
  • #physicalai
  • #aichip
  • #globalorders