NPU Neural Processor

NPUs are purpose-built for neural network inference, offering higher energy efficiency than GPUs, and are widely used for on-device AI acceleration. NPUs in mobile SoCs (such as Apple Neural Engine, Qualcomm Hexagon, MediaTek NeuroPilot, Samsung Gauss) handle camera enhancement, voice recognition, real-time translation, and AR scenarios with low-power inference, forming the core of the mobile AI experience. On the PC platform, AMD Ryzen AI and Intel AI Boost (NPU) enable laptops to run local large models without discrete graphics, suitable for privacy-sensitive productivity scenarios, playing a key role in the Windows 11 AI PC ecosystem. Independent NPU vendor Hailo focuses on edge computing and autonomous driving; the Hailo-8 is widely used in video analytics and industrial vision with excellent inference latency and power performance. The Huawei Ascend series covers everything from Atlas edge stations to data center training clusters, complemented by the MindSpore framework to build a full-stack AI infrastructure, holding significant market share in government, enterprise, and carrier markets. The Nexchip Dragon series targets smart cockpit and autonomous driving scenarios, filling the gap in domestic automotive AI chips.

This category includes the following AI accelerator chips/compute cards:

• Huawei Ascend
• AMD Ryzen AI NPU
• Intel NPU (AI Boost)
• Apple Neural Engine
• Qualcomm Hexagon NPU
• MediaTek NeuroPilot
• Samsung Gauss AI
• Hailo-8 / Hailo-15
• Nexchip Dragon Series