TLDR: DeepX, an on-device AI semiconductor company, has partnered with Samsung Electronics and Gaonchips to produce its next-generation generative AI chip, DX-M2, using Samsung’s advanced 2nm foundry process. This collaboration aims to significantly enhance power efficiency for demanding generative AI applications in resource-constrained environments, with mass production slated for 2027.
Seoul, South Korea – DeepX, a leading innovator in on-device artificial intelligence (AI) semiconductors, announced on August 13, 2025, a landmark agreement with Samsung Electronics and Gaonchips for the production of its cutting-edge generative AI semiconductor, the ‘DX-M2’. The chip will leverage Samsung’s highly anticipated 2nm (nanometer) foundry process, marking a significant step forward in the development of high-performance, power-efficient AI solutions.
The strategic partnership is set to revolutionize the deployment of generative AI models in various edge computing environments. DeepX anticipates that the utilization of the 2nm process will approximately double the power efficiency (performance per power consumption) of the DX-M2 compared to its predecessor, the DX-M1, which was manufactured using a 5nm process. This improvement is critical for generative AI, which demands substantial computational power, making its application in resource-limited on-device settings technically challenging. In this context, power efficiency emerges as the most vital metric determining a product’s practical viability.
The DX-M2 is engineered to perform real-time inference of generative AI models with up to 20 billion parameters, achieving speeds of 20-30 words per second while consuming less than 5W of power. This capability will enable expert-level AI models to operate autonomously through a device’s own computation, even in environments with severe thermal and power constraints, such as robotics, smart home appliances, and laptops. DeepX has been meticulously analyzing various aspects of Samsung Foundry’s 2nm process, including power efficiency, performance, manufacturing costs, and yield, since November of the previous year, confirming its ability to meet the stringent performance targets for on-device generative AI operations.
Prototype production for the DX-M2 is expected to commence in the first half of 2026, with mass production projected to begin in 2027. This initiative is viewed by industry experts as a pivotal development for the early establishment of the 2nm system semiconductor ecosystem and a significant opportunity for the domestic fabless industry to expand its high value-added semiconductor sector.
DeepX’s previous product, the DX-M1, developed two years prior on Samsung’s 5nm process, specialized in vision AI and is currently undergoing yield optimization. The DX-M1 is already being applied across diverse industries, including robotics, smart cameras, and factory automation, with collaborative development for mass production expanding with global clients. DeepX has also secured substantial funding, raising $531 million with backing from entities like Hyundai Robotics, underscoring investor confidence in its technological advancements.
Also Read:
- DeepX and Baidu Forge Strategic AI Alliance to Propel Edge Computing and Global AI Adoption
- SiMa.ai Unveils Modalix Chip for Energy-Efficient On-Device LLM and Physical AI Processing
Kim Nok-won, CEO of DeepX, emphasized the transformative potential of the new chip, stating, “DX-M2 is a core platform that opens the era of popularization and industrialization of generative AI technology.” He further added, “DeepX will continue to contribute to lowering technological barriers and creating a world where everyone can enjoy the benefits of AI.” This collaboration with Samsung is poised to accelerate the widespread adoption of sophisticated AI capabilities in everyday devices, making advanced AI more accessible and efficient.
