Google Coral Board Single-Board Computer Brings Local AI to Edge Devices

Google has unveiled the Coral Board, a single-board computer designed to bring local AI processing to developers and edge devices. The board features a custom NPU built on RISC-V architecture, enabling efficient machine learning tasks without relying on cloud connectivity.

Google unveils a new single-board computer featuring a custom NPU for local machine learning tasks at the I/O developer conference

The device centers around a Synaptics Astra SL2619 chip paired with Google's Coral NPU. It includes a 2GHz dual-core processor, 2GB of memory, and delivers 1 TOPS of compute performance for AI workloads.

Google demonstrated the board running the Gemma 3 model locally, targeting scenarios where network access is limited or privacy concerns are high. The system also handled real-time translation, voice-controlled hardware interactions, and generative music tasks during presentations at the I/O developer conference.

The Coral Board targets developers building AI applications for constrained environments. Its RISC-V-based NPU allows for efficient processing of machine learning models directly on the device, reducing latency and data transmission needs.

Google plans to release the board during the 2026 I/O developer conference. Specific pricing details have not been disclosed yet.