Andrew Hankins, head of engineering at Telstra Digital Infrastructure, and Sam Evans, senior managing director at Teneo, told the session, AI is changing network demand: What infrastructure teams need to do now, that most operators are still working from a pre-AI playbook even as traffic patterns shift beneath them.
Evans said that at an aggregate level, video still dominates network demand. But an “architectural effect” is already emerging, with sharper traffic peaks on specific data centre to data centre corridors and far less tolerance for latency or congestion.
Hankins gave a concrete example from Telstra’s own network: sustained AI traffic flows of hundreds of gigabits, running for hours over several days, originating from an Australian cloud operator and flowing to a US network Telstra had no direct interconnect with. The traffic reversed Telstra’s usual flow patterns entirely, forcing the company to work with the cloud operator to redimension interconnects it hadn’t planned for.
Both speakers pointed to agentic AI as the real inflection point. Unlike training workloads, agentic traffic is persistent and demands guaranteed, not just best-effort, latency, since a delay between communicating agents compounds rather than resolves. Evans argued this will push connectivity providers from selling infrastructure and services towards selling guaranteed outcomes.
Demand is currently concentrated in data centre to data centre corridors during the training phase, but both speakers expect it to move towards the edge as inference becomes more prevalent. Hankins said Telstra’s biggest concern is not core network dimensioning but interconnection, renegotiating capacity and commercial terms with other operators as traffic composition changes.
On bottlenecks, Evans and Hankins cited lengthening equipment delivery times, now around 12 months for some optical and IP hardware, subsea cable build times stretching to five years, and permitting delays across Asia-Pacific in particular. With roughly 60 cable-laying and repair ships operating worldwide, capacity for new builds is tightly constrained. Route security and sabotage risk were also flagged as growing concerns shaping where and how networks are built.
Both speakers agreed the industry is shifting towards capacity-as-a-service, using automation and pre-deployed, fungible capacity to cut provisioning times. Evans warned that operators who fail to adapt risk being reduced to commodity wholesalers in a value chain increasingly orchestrated by hyperscalers, facing rising volumes but falling unit prices and squeezed investment economics.
Watch the full webinar here.
RELATED STORIES:
