KPMG warns AI fiber connectivity demand faces power and labor bottlenecks

The shift from AI training to inference and agentic workloads creates a direct correlation between every gigawatt of new compute and a corresponding need for fiber capacity. For streaming and edge providers, this infrastructure strain threatens to increase transit costs and delay the rollout of localized AI services. As hyperscalers retreat from high-cost projects, the industry faces a fragmented connectivity map where ROI is no longer guaranteed in traditional business hubs. Watch for fiber operators to prioritize routes based on power grid stability rather than population density as they navigate these supply chain and labor shortages.

The AI-driven strain on network infrastructure is reshaping how major vendors architect their RAN and transport strategies. Ericsson launched its AI in RAN commercial software subscription on June 11, 2026, claiming up to 20% higher downlink throughput and 10% better spectral efficiency across more than 15 live deployments using existing baseband silicon. That same month, Nokia and Indosat Ooredoo Hutchison announced a GPU-accelerated AI-RAN partnership in Indonesia, expanding the Nokia-NVIDIA architecture already adopted by T-Mobile US, SoftBank, and Vodafone. These deployments underscore how AI workloads are forcing operators to rethink both compute placement and the fiber backhaul required to connect distributed inference nodes.

The competitive split between Ericsson and Nokia on AI-RAN architecture has direct implications for fiber connectivity demand. Ericsson and Nokia are diverging on whether L1 RAN functions should run on GPUs or CPUs, with Nokia committing its entire L1 stack to Nvidia's CUDA platform while Ericsson keeps only the FEC function on the GPU. Nokia's approach, backed by Nvidia's $1 billion investment, implies denser interconnect requirements between GPU clusters and the core network, potentially accelerating fiber buildout along those routes. T-Mobile US is testing both approaches, highlighting the split between incremental RAN intelligence and broader shared AI compute, a decision that will shape each operator's transport capacity planning for years.

On the operational side, agentic AI frameworks are emerging as a new category of network workload that will further stress fiber links. Nokia teamed up with Google Cloud to build six specialized Gemini-powered agents for autonomous network operations, targeting 50% to 80% reductions in problem-solving times, with a marketplace launch planned for September 2026. Verizon disclosed that its 60,000-site vRAN is now applying agentic AI to configuration changes and service assurance, while publicly calling for industry-wide interoperability standards. Ericsson's CTO Erik Ekudden , a forecast that directly reinforces KPMG's warning about fiber capacity constraints at the edge. This surge in demand is further evidenced by .