The role of fiber in AI scaling
As AI systems scale, network limitations increasingly inhibit performance, underutilize expensive GPUs, and reduce returns on large infrastructure investments.
Manish Rawat, semiconductor analyst at TechInsights, said fiber optics has emerged as the next structural constraint in scaling AI, which could have long-term implications.
“Fiber is a silent dependency that scales non-linearly as AI grows,” Rawat said. “AI workloads generate large east-west traffic and require tight synchronization across thousands of GPUs, resulting in exponential demand for light within data centers and between campuses.”
But so-called network walls are not the only bottleneck, said Sanchit Vir Gogia, principal analyst at Greyhound Research.
“This is a set of overlapping constraints that come to the fore as AI workloads scale, including fiber availability, switching density, optical transceiver limitations, and architectural inefficiencies,” Gogia said.
The scale of AI and the stress of parallel government broadband deployments have undermined the historical assumption that fiber is plentiful and cheap, Gogia added.
