When artificial intelligence (AI) models scale, they inevitably run into a hard physical limit. Modern data centers currently rely on traditional copper wiring to connect servers and graphics processing units. The fundamental problem is that copper generates intense heat, consumes substantial power and struggles with latency when pushing terabytes of data over long distances.
As frequencies increase to handle larger data loads, copper suffers from severe signal attenuation, meaning the data literally degrades in transit. The industry urgently needs a structural upgrade to the highway to keep pace with modern AI training algorithms.