On January 6, 2026, CES 2026 (Consumer Electronics Show) officially opened in Las Vegas, USA. As the annual global showcase for consumer electronics, the event highlights cutting-edge technologies and innovative products across AI, consumer electronics, display technology, robotics, and more.

From the scene at CES 2026, it is evident that AI remains the central theme. However, unlike CES 2025, with the continuous advancement of edge computing power and model performance, the AI industry is undergoing a profound paradigm shift. Application solutions are moving from a “training-focused” to an inference-focused approach, with large-scale model deployment becoming the main trend.

Amid this industry wave, MeiG Smart’s newly launched SNM983 high-performance AI module, delivering up to 100 TOPS, has emerged. This is not merely a performance upgrade; it represents an architectural revolution and design paradigm shift for the AI-native era. Its core feature is a complete transition from CPU-centric to scenario- and AI-centric design. Traditional SoC modules rely on general-purpose CPUs with AI as an auxiliary function, whereas the SNM983 is built from the ground up for high-complexity, low-latency edge AI inference, with all components designed to prioritize AI tasks while efficiently coordinating other units.

AI-Native: A High-Performance Module Built for AI Computing

Based on the Qualcomm IQ-9075 platform, the SNM983 flagship AI module features dual Qualcomm Hexagon™ tensor processors, delivering up to 100 TOPS of AI computing power. It also integrates a Kryo™ Gen 6 CPU (8×2.36 GHz), Adreno™ 663 @ 840 MHz GPU, and triple-channel LPDDR5 memory, combined with Qualcomm UWBC technology to achieve peak bandwidth up to 240 GB/s, providing high-bandwidth, low-latency support for intensive AI computing tasks.

The 100 TOPS performance comes from dedicated cores optimized for tensor operations, supporting sparsity and low-precision computing, forming an energy-efficient AI computing engine. In this architecture, the CPU evolves into an efficient scheduling and collaboration processor, creating a heterogeneous computing platform comparable to an “aircraft carrier”, designed to run large multimodal models at the edge in real time. This ensures deterministic performance for complex AI applications, such as real-time visual decision-making for autonomous robots, and provides a leading hardware foundation for AI algorithm development over the next 1–3 years.

Large Model Runtime Validation: A Critical Leap from Compute to Multi-Modal Capability

High-performance hardware is merely the foundation for LLM deployment. The real challenge lies in achieving efficient inference of multi-modal LLMs on resource-constrained edge devices. MeiG Smart has successfully completed on-device runtime validation of the QWEN2.5 7B multi-modal large language model on the SNM983 module, marking its capability to host state-of-the-art AI models.

This validation demonstrates that the SNM983 can support on-device inference for multi-modal and full-modal LLMs, directly deploying models with parameters ranging from 7B to tens of billions. It enables the joint understanding and reasoning of multi-modal input information (image, text, audio, etc.), completing complex on-device perception, comprehension, and decision-making loops.

The SNM983’s high computing power and multimodal inference capability make it suitable for a wide range of applications:

▶ AI Box & Edge Inference Devices: For multi-modal perception scenarios such as intelligent security, industrial quality inspection, and retail behavior analysis.

▶ In-Vehicle Multi-Modal Edge AI Computing Unit: Supports on-device processing of multi-modal and full-modal LLMs for automotive scenarios; enables fused processing of in-cabin visual perception, voice interaction, multi-screen infotainment, and driver status monitoring.

▶ Robotics & Unmanned Systems: Achieves integrated environmental understanding, voice command interaction, and navigation decision-making.

▶ Cloud-Edge Collaborative AI Node: Serves as an edge-side multi-modal signal processing unit, collaborating with cloud-based LLMs to form an efficient inference architecture.

▶ AR/VR Interaction Devices: Supports a fused experience integrating real-time visual positioning, gesture recognition, and voice interaction.

Outstanding Performance for Automotive Edge AI

At the system level, the SNM983 supports dual OS with Hypervisor (Ubuntu & Yocto Linux), offering flexible deployment in various embedded and edge computing environments, balancing development convenience and real-time requirements. For multimedia, it supports 4 camera inputs and features H.264/H.265 4K@240Hz decoding and 4K@120Hz encoding capabilities.

In terms of expandability and reliability, the SNM983 provides 2 RGMII+SGMII network interfaces, along with a rich set of peripheral interfaces including PCIe, I2C, I2S, SPI, UART, and DSC, covering mainstream communication and control protocols. Utilizing SIP (System-in-Package) technology, it achieves excellent thermal performance and environmental resilience within a compact form factor, making it suitable for harsh conditions in industrial, automotive, and outdoor applications.

Currently, an AI Box product equipped with the SNM983 module has secured a project designation from a renowned domestic automaker and will be deployed in its flagship models, providing high-realtime, multi-modal fused computing support for the intelligent cockpit. This solution supports in-vehicle multi-camera setups and microphone arrays, promising to enable more natural voice-visual fused interaction, adaptive in-cabin scene control, and immersive entertainment features.

It is also noteworthy that at CES 2026, the ceiling for edge AI performance was once again pushed higher, with Qualcomm releasing the new high-performance robotics processor — Qualcomm Dragonwing™ IQ10 series, featuring an 18-core Qualcomm Oryon CPU and up to 700 TOPS AI computing power, redefining the boundaries of edge AI. MeiG Smart will continue to strengthen its strategic partnership with Qualcomm, jointly driving the next generation of robotics, intelligent automotive, industrial IoT, and cross-industry large-scale innovation.

During CES, the SNM983 will be on display at the MeiG Smart booth (#10263, North Hall, Las Vegas Convention Center). We cordially invite you to join us.