AI does not thrive on compute alone. Compute is essential, but it only creates value when the system around it is trusted, resilient and connected.
This has been true for a while, and it is becoming harder to ignore as demand climbs. In early September, a working example showed how much room the network still has to grow. Telstra successfully transported four 400GbE client services over a single wavelength, without regeneration, operating in L-band spectrum between Melbourne and Sydney on a live production network running on Ciena's WaveLogic 6 Extreme (WL6e). The milestone extends the distance over which the companies previously demonstrated 1.6 Tb/s transmission by approximately 50% to approximately 1,100km, across one of Australia's busiest intercity network corridors. It was a world first: ultra-high-capacity wavelength transmission over the longest distance achieved to date using a single channel.
The technical gains behind it matter because of what they unlock. WaveLogic 6 Extreme delivers a 100% increase in per-channel capacity and approximately 14% increase in total system capacity. The addition of L-band further doubles usable spectrum, enabling up to two times more capacity per fibre. To optimise the use of optical spectrum beyond the traditional C-band, Telstra leveraged Ciena's 6500 Reconfigurable Line System (RLS) C+L band optical line system, together with WL6e. Telstra's Aura Network also made possible this landmark feat as a high-performing fibre route that delivers ultra-low loss infrastructure.
Capacity on that scale answers a question the industry has only recently started asking out loud. For a long time, the conversation in the AI industry was about compute: chips, racks, and GPUs. Increasingly, compute is not the constraint. AI infrastructure is intensely network-hungry. Training workloads move enormous volumes of data between GPU clusters, and inference needs to happen close to the user, which means moving data between core, edge and cloud. A hall full of powerful GPUs is an extraordinary concentration of capability, but capability only becomes useful when it can move. Connectivity is the part of the ecosystem most people do not see, and too often assume will simply be there.
That assumption is the risk. High-capacity routes exist today, terrestrially and in subsea, but many are aging or nearing capacity, and new routes take years to plan, permits and build. Projects like Telstra's Aura Network, connecting major capital cities and linking into international subsea systems, are being built in time for demand. As Telstra Digital Infrastructure Executive Sandy Cameron said, "This trial demonstrates how Telstra's Aura Network can unlock significantly more capacity by extending beyond the traditional C-band into L-band spectrum. For customers, that means greater scalability and the ability to support growing demand from AI, cloud, and data-intensive applications across Australia, both today and into the future."
Building ahead of demand is only part of the test. Resilience has to be designed in from day one, rather than added later. A year ago, quad diversity felt like the frontier. Today, the bar has moved. For the most critical workloads, customers are moving beyond dual, triple or quad diversity and asking for up to six physically separate routes across terrestrial and subsea, because they need confidence that no single cut, event, chokepoint or landing location can compromise the service.
There is a national dimension as well. Independent modelling by Oxford Economics estimates the committed Aura network alone could deliver $29 billion in additional GDP and support 84,000 extra Australian jobs by FY40, with the full rollout supporting $36 billion in GDP and more than 100,000 jobs nationwide.
Where AI thrives comes down to more than where data centres are built. It is the places where sovereignty, resilience and connectivity come together, and are built deliberately and early. A record transmission of over 1,100km, achieved without regeneration, on a route customers already rely on, is one answer to what that looks like.
