AI infrastructure faces power and fiber-optic bottlenecks

The rapid expansion of artificial intelligence is creating critical infrastructure bottlenecks in power supply and fiber-optic connectivity. While much focus remains on GPUs and memory, the physical requirements to link thousands of chips and power massive data centers are becoming primary constraints.

In terms of connectivity, traditional copper-based electrical connections face limitations in bandwidth and power efficiency as network scales increase. This is driving a shift toward optical communications. Technologies such as Co-Packaged Optics (CPO) and the transition of chip interfaces to optical signals are expected to reshape AI system architectures, potentially moving from integrated to flexible, distributed structures. Major players like NVIDIA and TSMC are already investing heavily in optical technologies.

Simultaneously, power availability is emerging as a decisive factor. Data center electricity consumption is growing rapidly, with projections suggesting demand could outpace total US power generation by the mid-2030s. Because power pricing is often determined by the marginal cost of the most expensive generator required to meet demand, the availability and efficiency of energy sources remain central to the scalability of AI infrastructure.