Kneron Edge AI NPU Chips and KNEO Platform

Kneron was founded in October 2015 in San Diego, California, by Albert Liu and develops reconfigurable neural processing units for on-device AI. Its goal is to let cars, cameras, IoT devices and small servers run AI locally, keeping data on site and reducing the cost and power of GPU-based inference.

The company offers a full stack: NPU system-on-chips and image processors, pre-built AI models, the KNEO software and runtime, KNEO Pi developer boards and edge servers. The NPU supports CNN and transformer networks, 2D and 3D vision and audio processing. The KL1140 chip, launched in November 2025, is described as the first NPU able to run full Mamba networks at the edge, and four cascaded chips can serve models of up to 120 billion parameters at one-third to half of GPU power. Kneron lists Toyota, Quanta, Hanwha and Dessmann among its customers and targets vehicles, AIoT, IP cameras, edge servers and education.

• Reconfigurable NPU: supports CNN and transformer networks for 2D and 3D vision and audio.
• KL1140 chip: runs full Mamba networks and brings language models to edge devices.
• Cascaded inference: four chips together serve models of up to 120 billion parameters.
• KNEO software: runtime and tools for deploying models on Kneron silicon.
• KNEO Pi: developer platform used by more than 28,000 developers worldwide.
• Edge servers: private on-premise AI servers built with partners such as Taiwan Spark Technology.

A Vietnamese enterprise can run a private document assistant on a Kneron-based edge server inside its office network. A smart lock, camera or appliance maker can embed Kneron NPUs for face recognition and voice control. A factory can analyse video and voice commands on the shop floor without relying on a cloud connection.

Kneron states that its NPUs deliver 6 to 10 times better power efficiency, and that the KL1140 offers up to 3 times greater energy efficiency and up to 10 times lower cost than current solutions for edge language models. Private assistants and video analytics can run on an office or factory edge server, keeping sensitive data on premises. Devices keep working without Wi-Fi or cellular links. A developer community of more than 28,000 KNEO Pi users eases prototyping.

Enterprises want generative AI assistants and video analytics without sending data to public clouds, which creates demand for small on-premise inference servers and AI-enabled devices. In Asia, camera, automotive and electronics manufacturers are adding local AI to products, and Southeast Asian firms with data residency concerns are a natural market for private edge AI.

Founded on 20 October 2015 in San Diego, California, by Albert Liu. Has raised more than 200 million USD from investors including Horizons Ventures, Qualcomm, Sequoia Capital, Alibaba Entrepreneurs Fund, Foxconn, CDIB and Himax. Employs more than 210 people, over 80% in R&D. Awards include the IEEE Corporate Innovation Award and COMPUTEX Best Choice Award.