- Bot traffic passed human traffic on the internet in May, according to Cloudflare
- Telcos need AI-enabled autonomous, secure and operationally sovereign networks to meet demand.
- The transition is well under way at leading telcos such as Verizon and Lumen
In a classic science fiction short story by Terry Bisson two aliens discuss a planet where the intelligent life forms are highly disturbing: "They're Made out of Meat." Their brains are made of meat, meat makes machines and their means of communication is downright disgusting:
"Meat sounds. You know how when you slap or flap meat, it makes a noise? They talk by flapping their meat at each other. They can even sing by squirting air through their meat."
The internet, on the other hand, is not made of meat. Or it's mostly not meat. Bots surpassed humans as the primary source of internet traffic in May, according to Cloudflare. Today, bot traffic now accounts for nearly two-thirds of internet traffic.
The shift to bot traffic — away from meat — is a problem for telcos, which built their networks, their staffing models and their capacity plans around people. The machines that now make up most of the internet's traffic don't sleep, don't take weekends off and don't call customer care when something breaks. They attack faster than humans can defend. And they raise a question: who is actually in control of the network?
To survive the shift, telcos need autonomous networks run by AI, AI-powered security to fight AI-powered attackers, operational sovereignty over their networks and the AI agents running them, and capacity for traffic that never lets up.
How do networks handle traffic that never sleeps
Human traffic rises and falls with the workday and peaks in the evening. Agents run around the clock. Cloudflare CEO Matthew Prince told The Verge that automated traffic could reach 1,000 times human traffic within five years.
Nokia expects machine-to-machine traffic to become the dominant driver of network growth.
Lumen is redesigning how connectivity is sold to match. Its Intelligent Internet service, launched in September, lets enterprises buy capacity in minutes through a portal or API and pay for what they use rather than for peak capacity.
Networks sized for evening peaks and quiet nights will need to absorb sustained, unpredictable load, and the only way to manage it is with automation that never sleeps either.
Why can't people run an agent-scale network
Human operators can't keep pace with machine traffic, and leading carriers acknowledge that reality.
Verizon ran 70 million closed-loop automations last year. Initially, those automations were scripted, but now Verizon is replacing scripts with reasoning.
Verizon's WatchTower agent system divides geography into small hexagons, watches network KPIs in each one and dispatches specialized subagents to correlate anomalies, find root causes and fix them. In a demonstration of a misbehaving cell in Midtown Manhattan, the system diagnosed a known software bug and reset the cell in 90 seconds. The same incident would have taken five to eight hours last year, Verizon CTO Yago Tenorio told Fierce in a June interview.
Cirion, which runs about 65,000 miles of fiber across Latin America, put an agentic AIOps platform into production in six months. In testing, it cut alarm noise by 90% to 92%, so a network operations center that once triaged 100 alerts could focus on about eight.
Telstra International is seeing why speed matters. Its AI customers buy capacity in terabits. Their "demands can change both in magnitude but also in location, so it's really raising the bar on us, and how quickly we're able to respond and turn up capacity," Regan Ireland, global head of pre-sales solution, products and digital experiences at Telstra International, told Fierce Network's Elizabeth Coyne in April.
What happens when attackers have agents too
Attackers adopted AI agents at least as fast as defenders did.
In November 2025, Anthropic disclosed that a Chinese state-sponsored group it calls GTG-1002 used Claude Code to run a cyberespionage campaign against about 30 organizations, with AI performing 80% to 90% of the work. In July, Palo Alto Networks' Unit 42 reported that Chinese-speaking operators used the DeepSeek model to autonomously find vulnerable systems, download exploit code and make more than 460 exploitation attempts.
Telcos are already prime targets. The Salt Typhoon intrusions into U.S. carriers were exposed in 2024, and in February an FBI official said the threat is "still very, very much ongoing."
The only defense that scales against machine-speed attacks is machine-speed defense. Lumen's Black Lotus Labs uses AI to analyze 200 billion NetFlow sessions and DNS queries every day and tracks 2.3 million threats daily.
Telcos also need to secure their own agents. An AI agent with permission to reconfigure the network is a high-value target for attackers.
Who controls the network's AI agents
Operational sovereignty means operators control their networks. That goes beyond where hardware, software and data physically sit, to who has authority over how the network runs. An operator can host every server in-country and still lose control if a foreign vendor, a cloud provider or a hostile government can push updates, hold the encryption keys or switch off a critical service.
"Sovereignty is not about a geographic place. Sovereignty is about control and where that control resides," my colleague Steve Saunders wrote in June.
AI agents raise the stakes. Network vendors now agree that operators must set the policies and permissions governing what agents can do. "Where a model or workload runs is not the same as who governs the network," said Akhil Gokul, head of the technology strategy office at Ericsson Americas.
Deutsche Telekom is putting sovereignty into practice for security. Its Sovereign Cortex with T Security offering, built with Palo Alto Networks, keeps encryption keys and telemetry under DT's control, uses independently audited access logs and staffs support exclusively with people based in Europe.
What will physical AI demand from telco networks
Bots browsing the web are just the beginning. Physical AI — robots, autonomous vehicles and AI-run factories, warehouses, cities, ports and shipyards — will put new demands on the network. And the stakes are higher: a dropped packet in a chatbot session is an annoyance, but a dropped packet in a welding robot or an autonomous crane is a safety incident.
Physical AI flips traffic patterns. Cameras and sensors push data up to the network, and Mobile Experts forecasts that uplink demand could outstrip capacity in 2029. Robots need predictable latency and coverage everywhere they work, not just where radio signals are strong. And they need inference nearby, because putting enough compute on every robot makes it unaffordable.
KT is building that network now. In September, KT deployed a 5G standalone network with network slicing, AI-RAN, GPU computing and autonomous network technology to run AI welding robots at the HD Hyundai Samho shipyard in South Korea. The network gives robots dedicated resources and monitors and adjusts itself in real time.
In the U.S., T-Mobile and Nvidia are working with Nokia to turn cell sites into edge AI platforms for smart city traffic management, drone-based utility inspection and industrial safety. "The radio tower used to be a radio tower. In the future, it's going to be an AI infrastructure platform," said Nvidia CEO Jensen Huang.
What should telcos do now
The message for operators is that a network built for people won't serve a world run by machines. Telcos need networks that drive themselves, govern themselves and defend themselves — and they need to keep their own hands on the controls while they do it.
The internet's new natives are machines. Telcos that build for the machines will carry the next era of traffic. The ones that don't will be left flapping their meat.
Read more about AI agents on Fierce Network
82% of rural broadband operators are stuck without an AI strategy