DeepX DX-M1: 20x AI Efficiency, 1/10th Cost; Next-Gen DX-M2 Targets Advanced Edge AI
South Korean AI semiconductor innovator, DeepX, is disrupting the burgeoning physical AI market with its mass-produced DX-M1 chip. This advanced AI chip offers a compelling low-power and cost-efficient alternative to Nvidia’s on-device GPUs, strategically positioning DeepX for significant growth.
DeepX highlights the DX-M1’s exceptional performance, operating at an average power consumption of just 2 to 3 watts. This translates to an impressive 20 times higher power efficiency compared to Nvidia’s Jetson Orin for similar workloads, all while costing approximately one-tenth as much. Furthermore, DeepX has successfully achieved production yields exceeding 90 percent for the DX-M1, a testament to its strong collaboration with foundry partner Samsung Electronics.
CEO Kim Lok-won emphasizes the crucial role of these advantages as AI applications expand beyond traditional data centers into diverse real-world environments.
“Just as power efficiency and heat control were pivotal during the shift from PCs to mobile computing, they are now critical for industrial and edge AI applications,” Kim stated during a press briefing in Pangyo on Tuesday. He added, “Even Nvidia, a dominant force in the AI chip market, encounters battery and thermal limitations in these demanding industrial and edge settings.”
Founded in 2018 by a former Apple chip designer, DeepX specializes in “physical AI.” This innovative field focuses on integrating AI directly into various devices, including advanced robotics, autonomous vehicles, and sophisticated factory automation systems.
DeepX proudly declares itself a “physical AI infrastructure company,” outlining an ambitious three-stage strategy. This comprehensive approach seamlessly connects cutting-edge AI chips with robust hardware platforms and expansive software ecosystems.
Under this strategic model, DeepX is focused on supplying its high-performance AI chips. Collaborating partners, including industry leaders like Advantech, Dell, and Raspberry Pi, are tasked with developing diverse hardware platforms. Concurrently, software partners such as Ultralytics and China’s Baidu are building innovative AI applications designed to run on these integrated systems.
DeepX is also actively addressing the challenge of transitioning from Nvidia-dominated GPU-based systems. Its innovative software platform, DXNN, facilitates the effortless execution of GPU-trained models on DeepX chips. Additionally, the DX-Newton interface significantly enhances compatibility with Nvidia’s widely used Isaac ROS robotics framework, streamlining integration for developers.
Kim projects substantial growth within the physical AI market, driven by the rapid expansion of robotics, the proliferation of smart factories, and the widespread adoption of AI-powered devices.
“With the accelerating spread of AI devices globally, the physical AI market is poised for more than threefold growth over the next five years,” he asserted. “DeepX is strategically positioned to build the essential computing infrastructure layer crucial for this dynamic market.”
The company also revealed its ambitious plans for the next-generation DX-M2 chip, which will leverage Samsung’s advanced 2-nanometer process technology. This groundbreaking chip is engineered to deliver an astounding 80 trillion operations per second (TOPS) while maintaining ultra-low power consumption of under 5 watts, with mass production slated for 2027.
DeepX confirmed that the DX-M2 chip is specifically designed to enable the deployment of large-scale AI models directly on edge devices. This innovation will effectively bring sophisticated AI capabilities, traditionally confined to data centers, into real-world applications across various sectors.
DeepX is actively preparing for an initial public offering (IPO), with a domestic listing currently considered the most probable path for the company.
Chief Financial Officer Cho Young announced that DeepX is targeting approximately $40 million in revenue for the current year, including $25 million from product sales. This aggressive target underscores the company’s commitment to building a strong commercial track record ahead of its anticipated IPO.
