Investment in artificial intelligence (AI) is reshaping mobile network requirements, forcing telecom operators to rethink traditional 5G investment priorities.
According to Ookla’s latest research, download speed alone is no longer enough to support AI applications such as ChatGPT, conversational AI, augmented reality (AR), AI-generated video, and agentic AI.
Instead, operators need to strengthen upload capacity, reduce latency, improve cloud connectivity, and optimize networks for always-on AI traffic. The report, based on Speedtest Intelligence data from 22 markets and 86 operators, outlines the investments required to prepare 5G networks for the AI era, the Ookla report said.
# Increase uplink capacity
Most 5G networks were designed around a 90/10 downlink-to-uplink traffic ratio, but AI applications require significantly more upstream bandwidth. Text-based AI already operates at around a 29/71 uplink-to-downlink split, while conversational AI approaches 50/50. AR and multimodal AI can require 40 percent or more uplink traffic. However, operators currently allocate only around 10 percent of network throughput to uplink, and fewer than half deliver the 20 Mbps upload speed needed for advanced AI services.
# Reduce latency across the network
AI applications increasingly depend on ultra-low latency rather than peak download speeds. Ookla recommends operators target below 50 milliseconds for text-based AI, below 40 milliseconds for conversational voice AI, and below 10 milliseconds for AR applications. While 18 of 22 markets already meet the text AI target and 13 markets meet conversational AI requirements, no market currently achieves the latency needed for immersive AR experiences.
# Improve network performance under congestion
Networks must maintain consistent responsiveness even during heavy traffic. Ookla found latency degradation under load ranges from 3.7x to 11.4x across markets. Operators should improve radio resource management, increase cell capacity, and optimize scheduling to prevent AI performance from deteriorating during peak usage.
# Deploy more 5G Standalone infrastructure
Emerging AI workloads such as conversational AI, industrial vision systems, and AR applications require 5G Standalone (SA) architecture, network slicing, and edge computing. These technologies provide dedicated low-latency paths for AI services, but remain unavailable across most commercial 5G deployments.
# Strengthen cloud connectivity
AI inference takes place largely on cloud platforms including Amazon Web Services (AWS), Microsoft Azure, Google Cloud, and Oracle Cloud Infrastructure (OCI). Ookla found cloud provider selection alone can change latency by nearly 100 milliseconds. Operators should therefore invest in better peering arrangements, shorter routing paths, and closer interconnection with hyperscale cloud providers.
# Prepare for AI traffic growth
AI-generated network traffic is forecast to expand at a 73 percent compound annual growth rate (CAGR) between 2025 and 2033, with AI traffic expected to exceed conventional mobile traffic by 2031. Operators must expand network capacity before demand reaches this tipping point.
# Support AI-capable devices
Operators should optimize networks for AI-enabled smartphones, smart glasses, industrial sensors, and wearables. Counterpoint Research expects AI smartphones to represent more than one-third of global smartphone shipments in 2025, while Omdia forecasts AI smart glasses shipments will grow from 5.1 million units in 2025 to more than 10 million in 2026 and 35 million by 2030, at a 47 percent CAGR.
# Prepare for enterprise AI workloads
Enterprise AI generates continuous upstream traffic from manufacturing, logistics, healthcare, and field operations. Cisco projects agentic AI could increase enterprise network traffic to approximately 9 times today’s levels by 2035, compared with about 2.5 times without AI. Meanwhile, Berg Insight estimates there were 6,500 private 5G deployments worldwide by the end of 2025, highlighting growing enterprise demand for high-performance mobile connectivity.
# Move beyond download speed as the primary KPI
Ookla recommends operators benchmark AI readiness using five core metrics: upload capacity, multi-server latency, loaded latency, cloud infrastructure latency, and jitter, rather than relying solely on download speed. These metrics better reflect real-world AI application performance.
# Build AI-ready networks for future applications
Future AI services including multimodal assistants, AI-generated video, conversational agents, and autonomous AI systems will demand continuous connectivity, higher upload bandwidth, and consistent low latency. Operators that rebalance network design around these AI-centric requirements will be better positioned to support the next generation of consumer and enterprise AI services.
BABURAJAN KIZHAKEDATH