For much of the semiconductor industry’s history, performance debates have centered on compute throughput and memory capacity. Faster processors, wider vectors, and larger caches have been the primary levers for system architects, often treating the movement of data between them as a secondary concern.
Today, a different constraint is asserting itself as AI data center workloads proliferate, architectures diversify, and systems extend beyond traditional computing into the physical world. Data movement, rather than processing or storage, is increasingly defining the limits of performance, power efficiency, predictability, determinism, and scalability.
