At a glance

  • Colocation held about 72 per cent of Australia’s data centre capacity in 2025, on one market estimate, while hyperscale self-build is the fastest-growing single segment.

  • Demand from global hyperscalers to lease colocation space now exceeds their own self-build programs, particularly in Sydney.

  • AWS has committed A$20 billion to Australian infrastructure across 2025 to 2029, Microsoft A$5 billion, while Google has paused a reported A$20 billion hub.

  • Around 78 per cent of Asia Pacific’s 4.8GW colocation development pipeline to 2027 is already pre-leased, much of it to hyperscalers.

  • Neoclouds and AI factories are a third model, renting wholesale capacity rather than building, which blurs the line between the two.


What hyperscale and colocation actually mean

Hyperscale and colocation describe who owns the building and who fills it. A hyperscale facility is built and operated by a single large cloud or technology company, such as AWS, Microsoft or Google, for its own workloads, usually at campus scale of hundreds of megawatts. Colocation is third-party space: an operator such as NEXTDC, CDC, AirTrunk, Equinix or Macquarie builds the facility and leases it to customers, either as multi-tenant retail space or as a dedicated wholesale hall.

A hyperscale self-build sits on the operator’s own balance sheet and gives it full control of design and timing. Colocation converts that into a lease, lets a customer scale without the upfront capital, and crucially can be energised in 12 to 18 months rather than the longer timeline of a ground-up campus. In a market where power and speed are the binding constraints, that difference is now decisive.

The market split: colocation still holds the base

By capacity, colocation remains the larger share of the Australian market. One market estimate put the colocation share at about 72 per cent in 2025, with hyperscale self-build recording the fastest growth rate of any segment. Both figures can be true at once, because the overall market is expanding quickly. Australian operational capacity is tracked at roughly 1.4GW today, heading toward 3.2GW by 2030 inside a much larger disclosed pipeline, as set out in our analysis of the national build-out.

The point is that this is not a zero-sum contest. Hyperscale capacity is growing fastest in percentage terms off a smaller base, while colocation continues to carry most of the installed footprint and is itself forecast to keep climbing as enterprises favour space that can be powered quickly.

Hyperscalers are colocation’s biggest customers

The framing that obscures the real story is “versus”. In practice the two models feed each other. Leasing demand from Western hyperscalers for colocation capacity now exceeds their own self-build programs in Australia, with Sydney the clearest example. Across Asia Pacific, about 78 per cent of the 4.8GW colocation pipeline to 2027 has already been pre-leased, a large part of it to the same hyperscalers that are also building their own campuses.

That is why the headline investment numbers cut both ways. AWS has committed A$20 billion to expand its Australian infrastructure in Sydney and Melbourne between 2025 and 2029, and Microsoft has committed A$5 billion to grow its footprint toward 29 locations, while Google has paused a reported A$20 billion hub. Those commitments fund self-build and leased capacity together. When a hyperscaler needs power in a hurry, it leases from a colocation operator, which is why NEXTDC and AirTrunk have built hyperscale-ready platforms designed to host exactly that demand.

Attribute

Hyperscale self-build

Colocation

Neocloud / GPU cloud

Who owns the building

The cloud operator

A third-party operator

Leases from a colocation operator

Who owns the servers

The cloud operator

The customer, or wholesale tenant

The neocloud, which owns the GPUs

Tenancy

Single tenant

Multi-tenant or wholesale

Wholesale tenant inside colocation

Typical scale

Hundreds of MW

Tens to hundreds of MW

Tens to hundreds of MW

Speed to energise

Longer, ground-up build

12 to 18 months

Fast, fitted into existing halls

Best suited to

At-scale cloud regions

Enterprise, government, and hyperscale leasing

AI training and inference

Australian examples

AWS, Microsoft, Google campuses

NEXTDC, CDC, AirTrunk, Equinix, Macquarie, DCI

Sharon AI, Firmus, ResetData

Source: Certified Strategic Editorial, operator disclosures and analyst reports, 2026.

Where neoclouds and AI factories fit

A third model has emerged since this comparison was first worth drawing. Neoclouds, or GPU-first cloud operators, do not build or sell general colocation. They lease wholesale space, fill it with their own GPUs, and sell compute by the hour for AI training and inference. Sharon AI runs its sovereign cluster inside NEXTDC, Firmus is building its own AI factories under Project Southgate, and ResetData offers sovereign GPU capacity. We set out this tier in full in our neocloud market report.

The effect is that the clean hyperscale-versus-colocation line has blurred into a stack. Colocation operators provide the shell and power, hyperscalers and neoclouds are the anchor tenants, and AI factories are a purpose-built variant optimised for dense, liquid-cooled GPU loads. For a fuller definition of that facility type, see our explainer on what an AI data centre is.

So what is winning

Both, and the more useful question is which model serves which need. Hyperscale self-build wins where a single operator needs control and scale for its own cloud region. Colocation wins on speed and flexibility, which is why it still holds most of the market and why hyperscalers themselves lease it. Neoclouds win the AI-compute layer that sits on top. The common thread is power: whoever can secure grid-connected megawatts fastest captures the demand, and colocation’s 12-to-18-month timeline is its decisive advantage in an AI cycle that will not wait.

For Australian enterprises, government and investors, the practical read is that the choice is rarely binary. Most large deployments now combine cloud regions, leased colocation and, increasingly, neocloud capacity. The directory of Australia’s data centre operators shows how concentrated the colocation base is and which providers carry the certified, government-ready facilities that sovereign workloads require.

What to watch

Three signals will show how the balance moves. Whether hyperscaler leasing keeps outpacing self-build as power constraints bite. Whether colocation pre-leasing stays near today’s highs, which would confirm capacity is spoken for years ahead. And whether the policy settings each state is putting in place, which we map in our guide to Australian data centre policy by state, shift where the next wave of capacity lands.