Gateworks Corporation and NXP Semiconductors Launch M.2 AI Accelerator

Gateworks Corporation and NXP Semiconductors have introduced the GW16168 M.2 AI acceleration card, a new edge-AI hardware solution designed around a decoupled AI architecture.

Unveiled at Embedded World 2026, the industrial-grade module integrates NXP’s Ara240 Discrete Neural Processing Unit (DNPU) and is designed, tested and assembled in the United States. The accelerator card aims to simplify AI deployment in embedded and industrial systems by allowing developers to add dedicated AI performance without redesigning the entire computing platform.

Modular AI acceleration for embedded systems
The GW16168 uses a standard M.2 2280 M-Key interface, allowing it to be integrated into existing embedded platforms and single-board computers. This approach enables developers to add AI processing capability without replacing their core computing hardware.

Gateworks designed the module based on a “Decoupled AI Architecture” philosophy, separating AI acceleration from the host compute platform. This approach allows system designers to upgrade AI performance independently of the main system hardware, improving flexibility and reducing long-term system redesign costs.

The card can be integrated with embedded processors such as NXP’s i.MX application processors, enabling AI workloads to run on the accelerator while the host CPU handles system logic, networking and I/O.

High performance with lower power consumption
The module features 16 GB of LPDDR4 memory and delivers up to 40 eTOPS of AI performance, enabling demanding workloads such as computer vision inference, large neural networks and transformer models.

Unlike GPU-based AI systems that often require significant cooling infrastructure, the Ara240 DNPU operates with passive cooling, helping maintain reliable operation in industrial environments.

Typical power consumption is approximately 6.6 W, enabling deployment in fanless or sealed systems where thermal constraints can limit traditional AI hardware.

Software ecosystem for edge AI deployment
The accelerator card is supported by the Ara240 SDK, which provides a full development environment for AI model deployment. The software stack includes:

This development environment simplifies the process of converting existing AI models for deployment on edge hardware.

Industrial design for long-term deployments
Gateworks built the module to industrial-grade standards, targeting applications in edge computing, automation, robotics, industrial IoT and smart infrastructure.

The company reports a projected lifespan of up to ten years, with thermal and power management designed to support long-term deployments in demanding environments.

The GW16168 AI accelerator card and associated development kit will be available through distributors including DigiKey, Avnet, Braemac and RoundSolutions, with shipments expected to begin in late May 2026.

Edited by an industrial journalist, Lekshman Ramdas.

www.gateworks.com