The firm seek to broaden scalable links between geographically distributed compute sites to cater to emerging “neoscalers”
To provide additional long-haul capacity between selected data center (DC) locations, Telstra is upgrading elements of its optical network in Australia.
The reason is that AI infrastructure providers are expanding distributed data center (DC) capacity to support model training, inference and related automation workloads. This creates demand for high-bandwidth, low-latency links between compute clusters, particularly where workloads and data must move across multiple facilities rather than remain within a single site. In Australia, the geographic separation of major data center hubs can make inter-DC connectivity a planning constraint. Operators must coordinate network capacity with DC construction, power availability and GPU deployments, while allowing for uncertain growth in east-west traffic between facilities.
The architectural improvement is intended to allow capacity increases without requiring operators to build and run their own long-distance optical networks. The use case is most relevant to AI-focused infrastructure operators — sometimes described by vendors as “neoscalers” — that need to establish or expand multiple compute sites quickly. Rather than owning all network infrastructure, these operators may use a carrier’s fiber routes and optical transport services while retaining control of their compute and data centre environments.
Said Sandy Cameron, the firm’s Infrastructure AI Executive: ““Unlike traditional enterprise customers, neoscalers operate highly dynamic environments where performance, latency, and rapid scalability are non-negotiable. AI workloads are driving exponential increases in inter-data center traffic, with capacity demands scaling to multiple terabits per route and requiring seamless, low-latency connectivity between compute clusters.”
The practical requirement for the upgrade is not simply headline bandwidth. Network redesign must address route diversity, latency between sites, resilience, provisioning lead times, interface compatibility and a credible upgrade path as traffic grows. Capacity requirements should therefore be assessed against actual training, storage replication and inference traffic patterns, rather than projected AI demand alone, according to the firm.
Tapping on external expertise to combining established fiber footprints, local delivery expertise, and investments in next-generation optical innovation, Telstra had contracted Ciena for the project.
