THE OPERATOR'S VIEW: INFRASTRUCTURE UNDER LOAD - TELECOMMUNICATIONS | ARTICLE 4 | 3-4 MIN READ
2 Article versions, the extended version is below the short version.
Data Centres: Comms as Data Infrastructure.
Australia's telecommunications network has changed buildings four times, and we are now seeing the next part in that evolution.
It started with stables, The Pony Express, with horses that carried mail between settlements. Then post offices, once the Postmaster General's Department strung telegraph wire and ran manual switchboards and party lines across the country. Then exchanges, purpose built for the PSTN's copper and coaxial cable. Telstra digitised those same exchanges through PDH, then SDH, then Ethernet over SDH, then IP. Each time, the building's job got bigger than the one it was built for, and each time unique requirements, of space, weight limits, fireproofing, power, heating and cooling. Good examples of industry adapted the building rather than starting over.
The current cycle is AI, the building this time is the data centre, increasingly carved directly out of the exchange real estate telecom already owns.
AEMO / AEMC GRID CONNECTION DATA, MARCH 2026 Eleven large scale data centre projects, representing 5.4 gigawatts of maximum demand, were progressing through the National Electricity Market's transmission connection process at the end of the March 2026 quarter, around 60 percent in NSW and 40 percent in Victoria. NSW alone has 44 data centres in its development pipeline totalling 11.4 gigawatts, close to four Eraring power stations. |
It stands to reason, much of that announced pipeline won't actually be built on the timeline promised. Industry estimates suggest only around 1.2 gigawatts of Sydney's pipeline will come online by 2030, a small fraction of the 11.4 gigawatts in the connection queue, because a meaningful share of applications are speculative, what the industry calls phantom demand. The grid is being asked to plan for a number that isn't real, while the compute that is real keeps landing on infrastructure never sized for it.
The building never stopped changing jobs. Only the equipment inside it kept getting bigger. |
Telstra's own exchange buildings show the pattern directly. Network Building Colocation lets carriers and resellers place equipment inside Telstra's fixed network exchange sites, across more than 100 interconnection points nationally. Telstra InfraCo now markets data centre colocation built explicitly for GPU processing and liquid cooling. The building that used to switch phone calls is, in places, now AI infrastructure.
The connective string is Telstra's Aura Network, a $1.6 billion national fibre backbone. Its Sydney to Melbourne route has demonstrated live speeds of 700 gigabits per second on a single fibre channel, with simulated capacity up to 83.6 terabits per second, around 35 times a standard carrier path. Microsoft is a foundation partner, using it to carry Azure and AI workloads nationwide.
Every generation says this is finally the finished network. In reality, as networks evolve, this is not true. |
None of this means the buildout is reckless, what it does mean, is that the pattern that has held since horseback mail, is happening once more: the network keeps getting asked to do more than the last version of it was built for, and the physical housing keeps having to catch up after the fact, not before.
In closing
Stables became post offices. Post offices became exchanges. Exchanges are becoming data centres. Each transition looked, at the time, like the final form of the network.
It never was, the grid connection queue for Sydney and Melbourne is the latest version of an old problem: demand for the network's next job is always easier to announce than the physical infrastructure is to build.
If you're interested in discussions on AI or how AI is actually delivered, across infrastructure, energy, networks, materials, and supply chains, subscribe: https://digitalbackbone-be8806.beehiiv.com/
Please see below for full article and more details.
THE OPERATOR'S VIEW: INFRASTRUCTURE UNDER LOAD - TELECOMMUNICATIONS | ARTICLE 4 | 6-8 MIN READ
Data Centres: Comms as Data Infrastructure.
Australia's telecommunications network has changed buildings four times. It's about to happen again.
Data centres tend to get discussed as something separate from telecommunications: a property and energy story, hyperscalers and colocation operators building warehouses full of servers, largely disconnected from the fibre and exchange network that carries a phone call or a broadband connection. The two are usually covered by different journalists, regulated by different bodies, and invested in by different parts of the market.
Australian telecommunications has changed its physical housing four times, and each time the trigger was the same: the job got bigger than the building. It started with stables, for the horses that carried mail between settlements before any wire existed. Then post offices, once the Postmaster-General's Department, established at Federation in 1901, strung telegraph wire and ran manual switchboards and party lines connecting the country. Telecom Australia took over in 1975, running the PSTN over copper and coaxial cable out of purpose-built telephone exchanges. Telstra, formed in 1993, digitised those same exchange buildings through successive transmission technologies, PDH, then SDH, then Ethernet over SDH, then Ethernet EVO, then IP, without ever needing a fifth kind of building. The exchange just kept absorbing bigger jobs.
The current cycle is AI, and the building this time is the data centre, increasingly carved directly out of the exchange real estate telecom already owns rather than built from scratch.
Where the system breaks
The break is in the mismatch between what's being promised on the grid and what the grid, or the fibre backbone feeding it, can actually deliver on the timeline being sold to the market.
AEMO QUARTERLY ENERGY DYNAMICS / CLIMATE COUNCIL, MARCH 2026 Eleven large-scale data centre projects, representing 5.4 gigawatts of maximum demand, were progressing through the National Electricity Market's transmission connection process at the end of the March 2026 quarter, around 60 percent in NSW and 40 percent in Victoria. NSW alone has 44 data centres in its development pipeline totalling 11.4 gigawatts, close to four Eraring power stations, though only an estimated 1.2 gigawatts is expected to actually come online in Sydney by 2030. |
That gap exists because a meaningful share of grid connection applications is speculative: requests for projects that will never be built, known in the industry as phantom demand, inflating the pipeline figures that make headlines. AEMO now treats data centres as a distinct planning category rather than a standard large industrial load, because their growth is faster, more concentrated, and less predictable than anything the grid has had to forecast before. The AEMC has drafted new technical standards requiring data centres over 30 megawatts to stay connected through grid faults rather than disconnect, after international incidents where mass simultaneous disconnections triggered cascading blackouts. NSW has gone further, legislating powers for its energy minister to control grid access for any load over 5 megawatts.
The building never stopped changing jobs. Only the equipment inside it kept getting bigger. |
Real world example
I started in this industry in Stirling Street, Perth, on the CAN2000 program and the 1998 IEN fibre rollout, laying the physical groundwork that would carry Western Australia's telecommunications for the next two decades. From there I moved to Sydney, working on SDH ring builds through major exchanges and the decommissioning of the old DCN-IP management network, including work at Dalley Exchange, a real Sydney CBD exchange near the harbour that had served the network since the early telephone era. Part of that work ran through the fibre tunnels connecting Sydney's exchange loop, physical infrastructure most people walking the CBD have no idea sits beneath them.
From Sydney I moved to Flinders Exchange in Adelaide, working through the transition to Ethernet over SDH and then Ethernet EVO, before the national NBN rollout took over most of the industry's attention for the best part of a decade. Most recently, my work has run through Project Aura, Telstra's national fibre program, on both the operating and capital expenditure sides of that build.
What stays with me across all of it is that no stage was ever described, at the time, as temporary. SDH felt permanent until Ethernet over SDH, then IP networks replaced it. The NBN rollout felt like the final infrastructure decision for a generation. Aura is being described the same way now, as the network built for the AI era. Every one of those descriptions was true for its moment and wrong about the future. The pattern isn't a flaw in any single decision. It's just what happens when a network built to carry one kind of traffic keeps getting asked to carry the next kind.
Every generation says this is finally the finished network. In reality, as networks evolve, this is not true |
Failure mode
What improves: AI and cloud workloads gain world class connectivity. Telstra's Aura Network has demonstrated live speeds of 700 Gbps on a single fibre channel between Sydney and Melbourne, with simulated capacity up to 83.6 Tbps, roughly 35 times a standard carrier route. Colocating compute directly inside legacy exchange buildings and interconnection points cuts latency and reuses infrastructure instead of building fresh from paddock. What degrades: The gap between what's being promised and what can actually be delivered. NSW alone has 44 data centres in its development pipeline totalling 11.4 gigawatts, close to four Eraring power stations, but industry estimates suggest only around 1.2 gigawatts of that will actually come online in Sydney by 2030. Exchange buildings and fibre routes upgraded within the last decade for broadband are now being asked to carry AI training and inference loads an order of magnitude beyond that design brief. Why it is not immediately visible: Data centre pipeline announcements are public and prominent. The gap between announced and deliverable capacity, much of it speculative connection requests known in the industry as phantom demand, only shows up when grid operators examine the connection queue in detail. The conversion of exchange buildings into compute infrastructure happens quietly, building by building, commercial deal by commercial deal, with no single moment where the shift from carrying calls to hosting AI gets publicly declared. Who owns the failure: AEMO and the AEMC own grid planning and the new technical standards now being drafted for large loads. NSW has legislated new powers to control grid access for big connections. Telstra and other carriers own the exchange buildings and interconnection points being repurposed, and the commercial pace of that expansion. No single body owns the question of whether a backbone built one generation ago for broadband can carry the AI compute economy on the timeline currently being promised to the market. |
Diagnostic table: what has housed Australia's telecommunications network, by era
Era | What carried it | What housed it | Primary task |
Colonial era (pre-1901) | Horseback riders, mail coaches | Stables, coaching inns | Physical mail delivery |
Postmaster-General's Dept (1901-1975) | Telegraph wire, manual switchboards, party lines | Post offices | Telegrams, basic voice |
Telecom Australia (1975-1993) | PSTN copper, coaxial cable | Telephone exchanges | Universal voice service |
Telstra, early digital (1993-2010s) | PDH, SDH, Ethernet over SDH | Exchanges, digitised with transmission gear | Voice plus early data |
NBN era (2010s-2020s) | IP, fibre (FTTP/HFC/FTTN), Ethernet EVO | Exchanges plus NBN points of interconnect | Broadband, video, cloud access |
Current (2020s-) | National fibre backbone (Aura Network), data centre interconnect | Data centres, AI-ready colocation inside former exchanges | AI compute, cloud, sovereign data |
Read across the table and the pattern is consistent across more than a century: the thing that carries the traffic changes, the building that houses it changes to match, and each era's operators genuinely believe they've built the final version. The AI era is not an exception to that pattern. It's the current instance of it.
What this means
For anyone assessing AI infrastructure investment or exposure, whether that's a data centre developer, an enterprise planning its cloud footprint, or an investor reading pipeline announcements at face value, the headline gigawatt figures in a state's data centre pipeline are not a reliable measure of what will actually be built and energised on the timeline claimed. The gap between announced and deliverable capacity is large, and it's structural, not incidental.
At small scale, colocating AI compute inside existing exchange buildings and interconnection points is the sensible, capital-efficient move, reusing infrastructure that already has power, cooling, and fibre connectivity rather than building greenfield. At system scale, the same decision means the AI compute economy is now riding on backbone infrastructure, exchange power capacity, and grid connections that were sized for the broadband and cloud era immediately before it, not for what's now being asked of them.
Demand for the network's next job is always easier to announce than the physical infrastructure is to build. |
Operator insight
Working through CAN2000, SDH ring builds, the NBN rollout, and now Project Aura, the one constant is that the technology transition always gets more attention than the physical building underneath it. Nobody writes headlines about exchange power capacity or fibre duct availability. They write headlines about the technology, IP, 5G, AI, that the building is being asked to carry.
But the building is the actual constraint, every time. Dalley Exchange couldn't run modern DCN-IP forever any more than a 1990s exchange could indefinitely support Ethernet EVO without work being done to it. The AI era's version of that constraint is grid capacity and data centre power, and right now the industry is doing what it has always done at this stage of a cycle: announcing the technology's potential faster than the physical infrastructure underneath it can actually be delivered.
What happens next
Expect the gap between announced and deliverable data centre capacity to become a bigger story than the AI compute story itself over the next few years, as AEMO's new planning categories, the AEMC's fault-ride-through standards, and NSW's grid access powers all work through the pipeline and separate the real projects from the phantom demand. Expect further exchange buildings to be repurposed for colocation and AI-ready capacity, following the pattern Telstra InfraCo has already set. And expect at least one more building transition after this one, because that's what the history in the table above says happens, without exception, once the current infrastructure gets treated as finished.
In closing
Stables became post offices. Post offices became exchanges. Exchanges are becoming data centres. Each transition looked, at the time, like the final form of the network.
It never was. The grid connection queue for Sydney and Melbourne is the latest version of an old problem, one I've watched play out from Stirling Street to Dalley Exchange to Flinders Exchange to Aura: demand for the network's next job is always easier to announce than the physical infrastructure is to build.
If you're interested in discussions on AI or how AI is actually delivered, across infrastructure, energy, networks, materials, and supply chains, subscribe: https://digitalbackbone-be8806.beehiiv.com/
Series navigation
Previous: Article 3, The Convergence: Circuit to Packet.
Next: Article 5, Space: Satellite and Sovereignty Risk.
References
AEMO: "Digital demand surge: Preparing Australia's power systems for the rise of data centres" (2026). Confirms the 11 large-scale data centre projects, 5.4 gigawatts of maximum demand, and the NSW/Victoria connection split at the end of the March 2026 quarter.
Climate Council: "Seizing the opportunity to do data centres right" (2026). Confirms the NSW pipeline of 44 data centres totalling 11.4 gigawatts, the Eraring comparison, and the estimate that only around 1.2 gigawatts will come online in Sydney by 2030.
Oxford Economics / AEMO: "Data Centre Energy Demand" final report (July 2025). Confirms current and forecast data centre electricity consumption as a share of NEM demand, and the distinction between grid-connected and behind-the-meter data centre load.
pv-tech: "Australia's data centre energy rules take shape as AEMC publishes framework and NSW moves to control grid access" (2026). Confirms the AEMC's fault-ride-through technical standards for large loads and NSW's new grid access legislation.
energy-storage.news: "Australia advances data centre energy policy on three fronts as battery storage emerges as key compliance tool" (2026). Confirms the NSW Electricity Infrastructure Investment Amendment Bill 2026 and its REZ-style grid access powers.
Telstra Wholesale: "InfraCo Building Access" (ongoing). Confirms the Network Building Colocation product allowing third-party equipment inside Telstra exchange sites across 100-plus interconnection points.
Telstra InfraCo: "Data Centre Colocation" brochure (ongoing). Confirms Telstra's own colocation offering built around GPU data processing and liquid cooling.
Telstra InfraCo: "Aura Network Trusted Digital Infrastructure, Built for AI" (2026). Confirms the network's 250,000-plus kilometre fibre footprint and connectivity to over 150 third-party data centres.
Telstra Exchange: "Telstra Aura Network" (October 2025). Confirms the live Sydney to Melbourne activation, the 700 Gbps demonstrated speed, the 83.6 Tbps simulated capacity, and Microsoft's role as foundation partner.
iTnews: "Telstra completes one-third of national fibre rollout" (2025). Confirms the Aura Network's $1.6 billion investment, the 14,000 kilometre build target, and the corridor-by-corridor rollout across Australian capital cities.
Whirlpool forums, PMG Exchange Photos archive. Confirms Dalley Exchange as a real, documented Sydney CBD telephone exchange, including its later partial decommissioning.
Wikipedia: "Victorian Telecommunications Museum." Confirms the 1975 split of the Postmaster-General's Department into Telecom Australia and Australia Post, and the 1993 formation of Telstra from the merger of Telecom Australia and OTC.
The views expressed in this article are my own and are intended for general information and discussion purposes only. They do not represent the views of any employer, organisation, or client.
© 2026 Rodney Terry, Digital Backbone. All rights reserved.