Apple, Xiaomi and Nvidia target faster AI chip data movement

Apple, Xiaomi and Nvidia are focusing on terabyte-class bandwidth as AI workloads shift the performance bottleneck in chips. The central challenge is no longer only making processors faster. It is moving enough data quickly enough to keep those processors supplied as AI demands rise.

For decades, semiconductor performance gains largely came from shrinking process nodes, moving from 22nm and 14nm to 7nm and 3nm. AI is changing that equation by making data movement an increasingly difficult constraint, putting bandwidth and the ability to feed high-performance processors at the center of chip development.

The shift underscores a broader change in semiconductor priorities for leading hardware companies. As processors continue to get faster, the ability to move data at terabyte-class speeds is becoming a defining issue for AI chip performance.