AI Traffic Tests Telecom’s Network Spending Story

Cisco forecast that AI could help push network traffic to 6.6 times current levels by 2035, but analysts and operator capex plans suggest many developed-market networks still have substantial unused capacity. Analysys Mason research cited busy-hour downlink loading of just 12% on older GPON broadband networks, while Omdia said annual RAN spending has stabilized at $35 billion after a $10 billion fall between 2022 and 2024. The key watchpoint is whether AI creates new supplementary traffic from vehicles, IoT and physical AI, or mostly substitutes for existing consumer data use without forcing a broad telecom spending cycle.

AI traffic is becoming a sharper test for telecom investment narratives as vendors warn about future network strain while operators and analysts point to large reserves of unused capacity.

A recent Cisco report forecast that network traffic could grow fourfold between 2025 and 2035 without AI, rise 6.6 times with AI, and leave AI responsible for a quarter of all traffic by 2035.

That framing matters because routers, switches, mobile radios and fiber upgrades are sold into expectations of demand growth.

The counter-signal is that many developed-market networks do not look capacity-constrained today.

Tom Rebbeck of Analysys Mason said fixed and mobile networks have the capacity customers need, and his research put average busy-hour downlink loading at just 12% for broadband networks based on the older GPON standard, with lower percentages on more advanced fiber and 5G networks.

Why it matters

The market question is not whether AI traffic grows.

It is whether that growth forces telecom operators to spend more, or simply changes how existing capacity is used.

That distinction matters for both operators and equipment suppliers.

Ericsson's mobile networks business saw organic sales growth fall to 1% last year from 4% when inventory buildup was a bigger topic in 2022.

Omdia analysts said annual radio access network spending stabilized at $35 billion after falling by $10 billion between 2022 and 2024, and Ericsson CEO Börje Ekholm said in January last year that the old hardware cycles are not expected to return in the same way.

Operators are also signaling tighter capital discipline.

BT plans £4.3 billion in spending this fiscal year, down from £5.1 billion in the prior year, as it approaches a fiber target of about 25 million premises.

Verizon plans $16 billion to $16.5 billion in capital expenditure this year, compared with $17 billion in 2025, while T-Mobile US has guided for $10 billion in 2026 after acquiring US Cellular in August 2025.

Who is affected

The immediate impact falls on telecom operators, network equipment vendors, fiber investors and enterprise customers watching the economics of AI connectivity.

If networks already have enough headroom, AI may not automatically produce a new spending cycle for suppliers.

If upstream-heavy AI use cases grow quickly, operators may still need targeted upgrades in busy areas.

AT&T shows the more nuanced case.

Its 2025 spending level was $22 billion, and management has set a $23 billion to $24 billion annual range for the following three years.

The increase is largely tied to fiber expansion from about 31 million premises today toward 60 million by 2030.

AI-related use cases could still change traffic direction.

At Network X in Dallas last month, AT&T network CTO Yigal Elbaz used the self-driving car as an example, noting that a vehicle can generate about 20 gigabytes of data per day, about 30 times an average mobile customer, with more traffic moving from the device to the network.

What to watch next

Readers should watch whether AI becomes supplementary network demand from vehicles, IoT systems and physical AI, or mostly substitutes for existing consumer activity such as video viewing and smartphone use.

William Webb, an academic, analyst and former regulatory executive, argued that fast early growth in AI traffic may slow as adoption levels off.

He also questioned broad extrapolation and said a more robust approach would count users, daily agent invocations and traffic per invocation.

The cautious conclusion is that AI may reshape telecom traffic patterns, but the investment case remains unproven.

For operators, the issue is targeted capacity and uplink readiness.

For suppliers, the harder question is whether AI creates measurable network spending or mainly becomes another story used to defend future demand.