China’s INN Inside, a brain-like supercomputer the size of a refrigerator, merges neural-style reasoning with machine learning to achieve GPU-level performance at minimal power. It signals a new direction in compact, energy-efficient AI computing.
China has unveiled the world’s first brain-like supercomputer named INN Inside, developed at the Guangdong-Macao In-Depth Cooperation Zone in Hengqin. The device, roughly the size of a household refrigerator, combines massive computing power with advanced neural-mimicking architecture. It is designed to simulate certain functions of the human brain, integrating learning and reasoning capabilities that allow it to process complex tasks in real time with exceptional efficiency.
The INN Inside system contains 1,152 CPU cores, 4.8 TB of memory, and 204 TB of storage. Despite this capacity, it consumes far less energy than traditional supercomputers, maintaining operation below 70 °C and generating under 45 decibels of noise. Its energy efficiency and compact design make it suitable for environments beyond data centers, such as laboratories, universities, and research facilities.
At the core of the system is a hybrid architecture that merges data-driven machine learning with symbolic reasoning, enabling the supercomputer to “think” and explain its analytical outcomes. During trials, it reached training speeds of about 100 K tokens per second and inference speeds of around 500 K tokens per second—performance levels comparable to GPU clusters. Chinese scientists believe this innovation marks a key step toward next-generation intelligent computing.