AI Semiconductor startup DeepX secured more than $13 million (about 18.4 billion won) in commercial purchase orders (POs) within a year of mass production with its first mass-produced product, the On-device AI semiconductor "DX-M1." More than 60% of the total orders were concentrated in the past four months.
DeepX said on the 14th that since starting DX-M1 mass production in Aug. last year through the end of July this year, it has secured a total of 77 purchase orders across more than 10 countries and regions, including Korea, the United States, China, Europe, Taiwan, Hong Kong, Japan, Singapore, and India.
DX-M1 is a Neural Processing Unit (NPU) for On-device AI that processes AI computation on the device itself without a consolidation to a cloud server. An NPU is a semiconductor designed to quickly process AI computations such as images and video. It is produced on Samsung Electronics' 5-nanometer (nm) process.
The pace of order growth has recently accelerated. Cumulative purchase orders increased from two in the third quarter of last year to six in the fourth quarter, 29 by the end of March this year, 56 by the end of May, and 77 by the end of July. Of the total 77, 48 orders (62%) occurred from April to July this year.
DeepX supplies DX-M1 to eight sectors, including robots and drones, smart factories, manufacturing automation, intelligent transportation systems (ITS), aviation, medical devices, and video control and security. The products are sold as chips and modules and include the DX-M1 series and some DX-H1 products.
According to the company, some customers have moved beyond supplying test products and technology verification to actual product development and production, followed by additional orders. As of the end of July this year, overseas sales accounted for more than half of cumulative revenue.
In China, Baidu-related optical character recognition (OCR) and machine vision projects and a Lenovo-related AI edge server project are underway. In Taiwan, the company is expanding product adoption with industrial computing company AAEON, and it is also supplying through local distribution networks and partners in North America, Europe, and Japan.
DeepX conducted sample product evaluations for more than 400 corporations for two to three years before mass production. After mass production, it signed contracts with 27 global semiconductor distributors and expanded its supply chain in cooperation with more than 50 hardware and module companies.
Next-generation product development is also underway. DeepX is developing "DX-M2," which applies Samsung Electronics' 2-nanometer process. Under the current plan, it will secure prototype wafers in Mar. 2027, verify chip operation and performance in Apr.–May, and begin preparing for initial customer supply starting in June.
DX-M2 aims to run Generative AI at up to 5 watts (W) of power per chip. It targets physical AI products with power constraints—such as robots and drones, industrial equipment, and battery-based smart devices—as its primary applications.
DeepX CEO Kim Nok-won said, "In the first year of mass production, actual orders and repeat orders occurred across multiple markets without being limited to a specific country or industry," and added, "We will expand our overseas business based on the mass production experience and global customer base we have built over the past year."