Phil Wong, KPMG US technology principal told RCR that fiber networks will face growing pressure as AI inference and agentic workloads create more traffic between cloud infrastructure, AI compute, and end users
In sum – what to know:
AI pushes fiber demand – Inference and agentic AI will drive demand for high-speed, low-latency connectivity, with traffic potentially moving closer to end users.
Power remains the constraint – Wong identifies reliable power as the primary limiting factor over the next three to five years, followed by supply-chain delays and labor availability.
Fiber ROI gets harder – New data center developments will require high-bandwidth middle-mile and, in some cases, long-haul routes, but operators will need to evaluate the ROI of individual deployments.
Fiber networks will face growing pressure as AI inference and agentic workloads create more traffic between cloud infrastructure, AI compute, and end users, according to Phil Wong, KPMG US technology principal.
Wong told RCR Wireless News that “traffic coming from inference workload will drive demand for high speed, low latency connectivity (i.e., fiber).” Agentic AI requires access to data, context and memory, increasing traffic between traditional cloud environments and AI-specific compute.
That demand could increasingly extend toward the network edge. “We could also see inference traffic spread more towards the edge of the network, closer to the end users, especially if physical AI takes off,” Wong said.
At the same time, AI infrastructure is expanding beyond traditional data center markets as developers confront land and power constraints. Wong said this will create demand for new high-bandwidth middle-mile and, in some cases, long-haul fiber routes to these locations.
“The challenge for fiber operators is whether they can get good ROI from these routes that may not pass through traditional population and business center,” he said.
While fiber requirements are increasing, Wong identifies reliable power as the primary constraint on scaling AI infrastructure over the next three to five years. “Currently, access to power, on or off-grid, is the biggest challenge, followed by supply chain delays and availability of labor,” he said.
Those challenges lead to longer deployment timelines and increased capital spending. Wong said that “in some cases, hyperscalers have cancelled already committed capacity because of the delays and prospect of ballooning costs.”
Power also sits at the center of the longer-term infrastructure equation. Wong said that most current capital expenditure is focused on GPU compute capacity, but connectivity requirements rise alongside new compute deployments.
“However, for every GW of new compute, there is a corresponding requirement for connectivity, and that would rise as workload shift from training to inference and agentic AI,” Wong said.
That creates a continuing connectivity requirement, but Wong said ROI for individual routes will need to be evaluated, particularly as some deployments move farther from traditional business and population centers.
Wong expects demand for AI-related infrastructure to continue growing in the near term as enterprises and consumers adopt AI and agentic AI applications.
“The demand of compute and storage is expected to continue to increase substantially as adoption of AI continues across enterprises and for consumers,” he said.
He also expects reasoning, multimodal processing, and physical AI to increase token consumption and, consequently, demand for compute and storage infrastructure.
“Agentic AI with reasoning, multi-modal processing, and physical AI are all going to drive explosion in token consumption and hence AI-related compute and storage infrastructure,” Wong said.
The interview with KPMG’s Phil Wong is part of a report published by RCR Wireless News and RCRTech, titled Scaling Optical Networks for the Hyperscale and AI Era, which can be accessed by clicking here.
