Edge artificial intelligence chipmaker Ambiq Micro Inc. took to the stage today at the 2026 Consumer Electronics Show in Las Vegas to unveil Atomiq, its first system-on-chip integrating a specialized neural processing unit for accelerating AI at the network edge.

Ambiq rose to prominence last year after launching an initial public offering in July that raised $96 million in funding. The company ended up selling a million more shares than expected and its stock gained more than 60% on its first day of trading, underscoring the huge enthusiasm among investors for its “edge AI” chips.

Its processors are designed for small, low-powered devices such as smart watches, drones and sensors. The chips can run powerful generative AI models locally, without sending data to the cloud to be processed.

Ambiq’s processors are based on a proprietary Subthreshold Power Optimized Technology or SPOT architecture that makes them five times more energy-efficient than traditional central processing units, making them ideal for battery-powered devices. One of its biggest customers is Samsung Electronics Co. Ltd., which uses its processors to power its Galaxy Watch devices.

With Atomiq, Ambiq is building on this foundation with the world’s first SPOT-optimized NPU, which promises additional efficiency gains. NPUs are specialized microprocessors designed to accelerate AI tasks by mimicking the processing functions of the human brain. They’re optimized for parallel processing and low-precision arithmetic, making them more efficient for AI applications compared to traditional CPUs and graphics processing units.

Atomiq is built on Arm Holdings Plc’s energy-efficient Ethos-U85 architecture and leverages SPOT’s frequency scaling capabilities to run at lower voltages, consuming even less power than before. With Atomiq, the company is promising superior levels of AI performance in a lower power budget than competing NPUs.

Ambiq says Atomiq’s advances can help manufacturers to scale on-device AI intelligence without compromising battery life or making their devices run hotter. The chips have been architected to strike the perfect balance between compute performance and power consumption to support sustained AI processing at the edge. Ambiq claims it can perform up to 200 billion operations per second, making it fast enough to run AI workloads such as computer vision, multilingual speech recognition and sensory models locally on battery-powered devices.

Ambiq said the Atomiq NPUs are ideal for platforms including smart security cameras, augmented reality glasses and wearable devices featuring high-resolution object recognition, advanced real-time tracking capabilities and conversational voice interfaces. They can also be used in industrial applications and robotics that need to perform high-fidelity semantic audio processing and data analysis on factory floors. Developers can take advantage of the company’s Helia AI Platform and its AI development kits or ADKs and the modular neuralSPOT SDK to customize the processors for specific devices and applications.

“The Atomiq product family represents a step-change in energy-efficient edge AI,” said Ambiq founder and Chief Technology Officer Scott Hanson. “By combining the Atomiq system architecture with Arm Ethos-U85, we enable significantly larger AI models at the edge with industry-leading energy efficiency.”

The company believes it will be able to deliver further energy-efficiency gains in future. At CES it revealed a roadmap for shifting its SPOT platform to a new 12-nanometer architecture that will be able to support more sophisticated AI applications at the edge, including new use cases in healthcare, consumer electronics and smart buildings.

Photo: Ambiq