Edge-AI chipmaker DeepX’s DX-M1 neural processing unit entered mass production using Samsung’s 5-nanometer process; 2-nanometer tape-out remains planned this month.

DeepX’s current production milestone concerns the DX-M1 neural processing unit, or NPU. The South Korean edge-AI startup is reported to have achieved mass production using Samsung’s 5-nanometer process. Its separate 2-nanometer chip remains at the planned tape-out stage, with tape-out targeted for October and first chips expected in March–April 2027.

Software integration and partnerships are at different stages. DeepX is working to integrate its NPU into Intel’s OpenVINO ecosystem, with results expected in early 2027, and is discussing physical-AI cooperation with AMD. The report also identifies Samsung Electronics and Hyundai Motor Group among the companies with which DeepX is expanding cooperation, without detailing a Hyundai agreement.

The source provides order and distribution figures, but no revenue totals. Between August 2025 and July 2026, DeepX received 77 orders from more than ten countries, signed 27 global semiconductor distributors, and had approximately 300 organizations evaluating its products. Order values, chip quantities, and shipment schedules were not provided.

Expansion and listing remain management plans. DeepX plans branches in Japan and Europe and intends to release DX-Newton robot-management software compatible with NVIDIA’s Isaac robotics platform later in 2026. Chief executive Kim Nok-won said he hopes to complete an initial public offering within three years; the report does not describe a filed offering, financing round, valuation, or offering terms.