Optical networks key to AI data center growth, says NTT
NTT Global Data Centers' Bruno Berti says AI workloads demand flatter, more deterministic networks with higher-speed optics and greater fiber density. Optical technology can extend existing fiber but will not eliminate the need for additional fiber as AI traffic scales. Integration of compute, storage, and networking becomes increasingly critical in data center design.
AI workloads differ sharply from conventional cloud traffic, which flows predictably between users and applications. Training and large-scale inference instead generate enormous internal data movement as GPUs communicate constantly, with latency sensitivity measured in microseconds. This demands flatter, more deterministic network architectures with higher-capacity fabrics and advanced optical interconnects, according to NTT's Bruno Berti.
Higher-speed optics—including 400G, 800G, and 1.6T—can extend the useful life of existing fiber at data-center and metro levels, but Berti stressed they cannot eliminate the need for targeted new fiber deployment as AI scales. He emphasized that network planning must proceed alongside compute expansion, warning that bottlenecks emerge when networking is treated as an afterthought rather than core AI infrastructure.
This shift could affect enterprises and consumers relying on AI services, as data center design directly influences the cost, speed, and reliability of AI applications. If optical innovation delays expensive fiber builds, providers may pass savings to customers; conversely, targeted expansion costs could raise infrastructure spending. Communities hosting data centers may see altered construction patterns, while businesses dependent on AI inference could experience performance variations tied to network architecture decisions.