Fabless semiconductor firm Ambient Scientific has introduced the GPX10 Pro, an AI-native system-on-chip (SoC) designed to deliver high-performance artificial intelligence inference on battery-powered edge devices. The launch marks a significant step in enabling ultra-low-power applications such as keyword spotting, facial recognition, and intelligent sensing.
The GPX10 Pro is built on the company’s proprietary DigAn® architecture, which maps neural network matrix-multiply operations and activation flows directly onto in-memory analog compute blocks. By removing the inefficiencies of conventional instruction sets, the chip offers up to 100 times improvements in power, performance, and area compared with standard 32-bit microcontrollers.
“Today’s MCUs and NPUs are hamstrung by their conventional silicon architecture when they try to run AI models. It’s like hitting a baseball with a tennis racket – the wrong tool for the job,” said GP Singh, chief executive officer of Ambient Scientific. “The GPX10 Pro shows what’s possible when you build your architecture natively for AI – hundreds of GOPs of AI performance at microwatts of power.”
The GPX10 Pro integrates 10 programmable MX8 AI cores, split into two power domains. One domain includes an always-on block designed for ultra-low-power sensor interfacing and fusion. For example, keyword spotting tasks consume less than 100 microwatts. Together, the cores deliver up to 2,560 multiply-accumulate operations per cycle, with peak throughput of 512 GOPs.
The new processor includes 2MB of on-chip SRAM, ten times more than the earlier GPX10, enabling larger and more sophisticated AI models. Supporting logic includes an Arm® Cortex®-M4F CPU for control functions, while integrated analog features extend to a low-power ADC, enhanced I2S logic, and interfaces for up to eight analog and 20 digital sensors.
The company also unveiled tools and software to ease adoption. Its Nebula™ AI toolchain supports training and deployment using widely used frameworks such as TensorFlow, Keras, and ONNX, while providing programmability across AI cores to adapt to evolving model topologies. Complementing this is the SenseMesh™ hardware sensor fusion layer, which connects multiple sensors through a tightly-coupled mesh to enable low-latency responses and reduced CPU workloads.
The launch underscores a broader industry shift toward energy-efficient AI at the edge. With demand for devices that can process data locally without relying on cloud infrastructure, chipmakers are racing to build architectures that deliver real-time inference while conserving battery life.
Ambient Scientific is showcasing the GPX10 Pro at the Electronica exhibition in Bangalore from September 17-19, demonstrating on-device AI applications such as fall detection, voice recognition, and face identification powered by coin cell batteries.
The GPX10 Pro is currently available for sampling, with volume production slated to begin in the first quarter of 2026.
