Australia's Next Infrastructure Race Isn't About Chips

We spend our days identifying land with spare grid capacity, running it through the network connection process, and walking it to build-ready, first for batteries and solar, now for data centres. It's about land and power, and it looks a lot like a market we already know.

As of March 2026, developers had proposed 21.6 gigawatts worth of new data centres across Australia, according to the Australian Data Centre Forecast Report [1]. To put that in perspective: a gigawatt is roughly enough electricity to power a mid-sized city. And the capacity that's actually up and running is expected to more than double too, from 1.4 gigawatts in 2025 to 3.2 gigawatts by 2030 [1]. This isn't hype. It's a construction pipeline bigger than the entire grid of some Australian states. Everyone has noticed the AI headlines. Fewer people are asking the more practical question: who is actually going to build the thousands of smaller sites this boom will need, and where? That's exactly the question our development pipeline already answers for batteries and solar, and now for data centres too.

Three forces pulling compute onto Australian soil

Start with private enterprise. Australian banks, telcos, retailers, and other large corporates are increasingly building or leasing dedicated AI infrastructure rather than relying purely on public cloud, driven by a mix of latency, reliability, and customer trust. Keeping sensitive customer data onshore, rather than routed through an overseas cloud region, has become a competitive differentiator and increasingly shows up as an explicit sourcing requirement that large corporates place on their own technology vendors and partners.

The demand for this isn't theoretical. CommBank built an "AI Factory" with AWS and became the first Australian customer to use AWS's newest AI computer chips (EC2 P5) here in Sydney, giving the bank dedicated computing power to train and fine-tune its own AI models [2]. ANZ opened an AI training centre with Microsoft at its Melbourne head office, backed by 3,000 staff licences for Microsoft's AI assistant, Copilot [3]. Telstra InfraCo is going even further, partnering with Starwood Capital and Doma Infrastructure to build a 62 megawatt data centre purpose-built for AI in Western Sydney [4]. Running AI at scale isn't a cloud subscription. It's a physical build, and it needs to be close, secure, and reliable.

Independent forecasters expect this to keep accelerating. Research firm Mordor Intelligence expects Australia's large-scale data centre market to grow from around US$6.3 billion in 2026 to over US$16 billion by 2031 [5]. JLL expects total data centre capacity across the Asia-Pacific region to grow from 32 gigawatts to 57 gigawatts by 2030 [6]. Property firm CBRE expects the big cloud companies (the "hyperscalers") to spend more than US$400 billion building data centres across the Asia-Pacific region in 2026 alone [7]. These are estimates from commercial research firms, not guaranteed outcomes, but every one of them points in the same direction.

Australia's operational data centre capacity is set to more than double by 2030

Forecast operational capacity, gigawatts (GW), 2025 to 2030

21.6 GW

already proposed

Nearly 7× the capacity forecast to actually be operating by 2030. Most of this pipeline still needs land, grid connections, and approvals before it's real.

Operational capacity is forecast to grow from 1.4 GW in 2025 to 3.2 GW in 2030 (interpolated trend shown); intermediate years are illustrative. Source: Australian Data Centre Forecast Report, April 2026 [1].

The constraint has moved

For years, the limiting factor for a new data centre was floor space: finding a big enough building or block of land. That's no longer the bottleneck. Today, the real constraint is grid power: whether the electricity network actually has enough spare capacity to plug the data centre in. The market operator's own numbers show why. As of the end of March 2026, eleven large data centre projects (each over 5 megawatts) needing a combined 5.4 gigawatts of power were sitting in the queue for a transmission grid connection, split roughly 60/40 between New South Wales and Victoria [8]. The market operator (AEMO) says these large connections are currently taking around two years from application to actually being switched on [8], and has also flagged that data centres need careful management to avoid destabilising the grid's voltage and overall stability, especially where several projects are connecting in the same area without coordination [8].

The queue for grid power is longest and land is scarcest in the biggest capital city markets [9]. That's pushing serious development activity outward too, into outer suburbs and regional centres where the grid still has room to spare, opening up a gap alongside the metro markets that the biggest, well-capitalised operators have less incentive to fill themselves.

Not every data centre is a hyperscale campus

It helps to be clear about what "data centre" actually means, because in practice the term covers three quite different kinds of business, at very different scales.

20 MW AND ABOVE

Hyperscale

These are the massive campuses built by Amazon, Microsoft, and Google, the kind that make headlines. They function as the backbone of global cloud computing: training huge AI models, running search engines, and hosting the software millions of people use every day. They're billion-dollar builds that take years to plan and construct, need enormous, dedicated power supplies (often their own substation), and are usually only viable for the world's largest tech companies. This is not where a smaller, nimble developer can realistically compete.

1-5 MW · OUR FOCUS

Metro and regional edge

These are smaller, purpose-built facilities (sometimes even assembled from shipping-container-style modules) that serve a single city precinct or regional hub rather than the whole country. Functionally, they do the jobs that need to happen close to where people actually are: running AI "inference" (using an already-trained AI model to answer a real query quickly, as opposed to the much heavier job of training the model in the first place), supporting 5G mobile networks, and hosting sensitive data that banks, healthcare providers, and other enterprises increasingly want to keep onshore [10]. Because they're smaller, they can be built faster and sited more flexibly than a hyperscale campus. This is the gap the big players aren't filling, and it's where we focus.

UNDER 0.5 MW

Micro edge

These are single, small container-sized units placed directly at a hospital, a factory, or a phone tower. They function as very localised computing power: for example, processing medical scans on-site at a hospital, or running factory automation systems in real time, without needing to send that data anywhere else. They're often built to run off-grid or on a hybrid mix of grid and on-site power (like solar and batteries), and are deliberately spread out across many small sites rather than centralised in one place.

We're not chasing the hyperscale race. We're focused on the metro and regional edge tier, because it's where the site-finding and development skills we've already built for battery and solar projects translate most directly.

Why this looks familiar to us

Developing a data centre connected to the local electricity network follows almost exactly the same steps as developing a small-scale battery or solar project. First, you identify land with the right characteristics: spare grid capacity, road access, the right zoning, and proximity to the people or businesses that will use it. Then you lodge a grid connection enquiry and work through the network operator's process to understand exactly what the site can actually support. Alongside that, you work through planning approval (whether that's a straightforward council permit or a faster, facilitated pathway) together with council and community engagement. Throughout, you're resolving the technical, environmental, and regulatory unknowns before anyone commits real money to construction.

The end customer is different. A battery's customer is the wholesale energy market. A data centre's customer is a telco, a corporate enterprise, or a business with a computing need to solve. But the actual discipline of getting a site from raw land to "ready to build" is structurally the same process either way.

Our geospatial screening tool already maps land with spare grid capacity across New South Wales, Victoria, and South Australia, the exact same information that matters for choosing a data centre site. We already have hands-on experience working through network connection processes and planning approvals, and we already assess sites against zoning, grid proximity, environmental constraints, land ownership, and connectivity: precisely the filters an edge data centre site needs. This isn't a stretch for us. It's close to home.

Where we think this goes

AI data centres in Australia today look a lot like utility-scale batteries did before the Hornsdale Power Reserve was built. The demand signals are already this clear, and most of the capital hasn't moved yet.

South Australia's "big battery" (Hornsdale Power Reserve) switched on at 100 megawatts in November 2017 (the first of its kind in the world) and grew to 150 megawatts within three years [11]. Before that, there was barely a market for grid-scale batteries in Australia at all. Within a few years, batteries went from a novelty to a multi-gigawatt national pipeline. We think edge data centres are at a similar starting point today: the demand signals are already this clear, the project pipeline is growing fast, the rules are still being written, and most of the capital hasn't moved yet. The developers who build up site-finding capability and local knowledge now, before good sites become scarce, are the ones who will capture the value as this cycle plays out. We're building that capability now. For us, it's an extension of what we already do, not a pivot into something new.

Let's talk

We're actively finding edge data centre sites and talking with developers, operators, and investors who are thinking through their own data centre strategy. If that's you, we'd welcome the conversation. Get in touch through the OnyX website or send us a message on LinkedIn.