DEEPX, a pioneer in ultra-low-power on-device AI semiconductors, today unveiled its next-generation roadmap at a global media briefing during CES 2026. Headlining the announcement is the DX-M2, the company’s second-generation AI processor designed to run Large Language Models (LLMs) of up to 100 billion parameters at under 5 watts—a breakthrough set to liberate generative AI from the constraints of cloud data centers.

DX-M2: Bringing Generative AI to the Edge (Sub-5W) The DX-M2 roadmap targets on-device inference for 20B–100B parameter LLMs, including Mixture-of-Experts (MoE) architectures, within a remarkably low power envelope of sub-5W. This represents a paradigm shift, migrating generative AI workloads from centralized servers to independent edge devices operating in the physical world.

“Running a 100B parameter LLM under 5 watts isn’t just a performance upgrade—it redefines the fundamental possibilities of AI infrastructure,” said Young Cho, Vice President of DEEPX. “Generative AI will no longer be tethered to the cloud. It will be ubiquitous, embedded directly into the machines that perceive and interact with our world.”

This vision positions the DX-M2 as a critical enabler for “Physical AI”—empowering real-world machines to process, interpret, and act autonomously without reliance on network connectivity or energy-intensive server farms.

Impact on Infrastructure & Sustainability DEEPX forecasts that distributing inference workloads across billions of edge devices could slash data center traffic by over 80%, offering a viable solution to current energy bottlenecks and escalating operational costs. The DX-M2 leverages architectural insights from the mass production of the first-generation DX-M1, featuring enhanced power management, superior thermal efficiency, and optimized software stacks.

Scaling the Ecosystem: 3 Strategic Pillars Building on the commercial success of the DX-M1 platform—now deployed in over 50 projects across robotics, defense, and industrial sectors—DEEPX introduced a three-pillar strategy to accelerate global adoption:

1. Open-Source Physical AI Alliance

Partners: Ultralytics, Baidu PaddlePaddle

DEEPX is establishing a developer-centric software ecosystem tailored for real-world deployment. By integrating with leading frameworks like Ultralytics and Baidu PaddlePaddle, DEEPX ensures a seamless workflow from model development to edge deployment, minimizing friction for engineers.

2. Global Distribution Network

Partners: Avnet, DigiKey, WPG

To streamline supply chains and improve accessibility, DEEPX has secured partnerships with major global distributors including Avnet, DigiKey, and WPG. Chips, development kits, and reference modules are now readily available, accelerating time-to-market for customers worldwide.

3. “AI Player” Dev-Kit

The newly launched AI Player Dev-Kit lowers the barrier to entry for edge AI. Designed to simplify model evaluation, benchmarking, and system validation, this tool shortens the cycle from prototyping to mass production, abstracting hardware complexity for developers and integrators.

“While the DX-M1 proved the viability of large-scale vision AI at the edge, the DX-M2 marks the inflection point for generative AI,” a DEEPX spokesperson noted. “Our mission is to decouple AI from the cost and energy limitations of traditional centralized infrastructure.”