How is an AI data centre different from a normal data centre?

The core difference is power density. A traditional server rack draws around 5 to 10 kilowatts, while an AI rack such as NVIDIA’s GB200 NVL72 is rated above 120kW. That concentration of power produces heat that air cooling cannot handle, so AI data centres adopt direct liquid cooling. It also changes site selection: because the facilities need so much electricity, they are increasingly built near grid capacity and renewable generation rather than only near cities and fibre.

Why do AI data centres need GPUs?

GPUs, or graphics processing units, perform thousands of calculations in parallel, which is exactly what training and running AI models requires. A CPU handles tasks in sequence and is far slower for this kind of maths. An AI data centre wires thousands of GPUs together so they operate as a single large machine, a configuration the industry calls an AI factory. The scale of GPU clusters is the main reason AI data centres consume so much more power than conventional facilities.

What is an AI data centre? | Australia Data Centre Index

Q: What is an AI data centre?

An AI data centre is a facility purpose-built to train and run artificial-intelligence models. It is filled with high-density GPU servers, uses liquid cooling to manage the heat, and draws far more electrical power per rack than a conventional data centre. While a traditional facility mainly stores data and runs cloud applications, an AI data centre exists to perform the large-scale parallel computation that AI training and inference require. In Australia, a growing share of new and retrofitted capacity around Sydney and Melbourne is being built for these AI workloads.

Q: How many AI data centres are there in Australia?

Certified Strategic tracks 208 data centre sites across Australia, a growing share of which are being built or retrofitted for AI workloads, concentrated around Sydney and Melbourne. A subset are government-approved and certified under the Hosting Certification Framework for hosting sensitive workloads. The full, searchable and filterable list, including certification status, is maintained in the Certified Strategic directory of data centres in Australia.

Defining an AI data centre

An AI data centre is a building full of computers optimised for one job: training and running artificial-intelligence models. It looks like a traditional data centre from the outside, racks of servers, backup power, cooling, security, but the machines inside and the power feeding them are a different class.

A conventional data centre stores files, runs websites, hosts company software and serves cloud applications. An AI data centre does that style of work too, but its core purpose is to crunch the enormous parallel calculations that large AI models require, both to train them in the first place and to answer queries once they are live.

How AI data centres differ from traditional data centres

The clearest way to see the difference is power per rack. A standard enterprise rack draws roughly 5 to 10 kilowatts. An AI rack built around the latest accelerators draws far more: NVIDIA rates its GB200 NVL72 system above 120kW in a single rack. Concentrating that much electricity in one cabinet produces heat that ordinary air conditioning cannot remove, so AI facilities move to direct liquid cooling, piping coolant to the chips themselves.

Attribute	Traditional data centre	AI data centre
Primary purpose	Storage, web hosting, cloud applications	Training and running AI models
Power per rack	Around 5 to 10kW	120kW or more (NVIDIA GB200 NVL72)
Cooling	Air conditioning	Direct liquid cooling
Core hardware	CPU servers	GPU clusters, or “AI factories”
Main siting driver	Proximity to cities and fibre	Access to grid power and generation

Source: Certified Strategic Editorial, NVIDIA system specifications, June 2026.

Higher density also reshapes where these facilities go and how they connect. Traditional data centres cluster near cities and fibre. AI data centres still need fibre, but their appetite for electricity means proximity to grid capacity and generation now drives site selection as much as connectivity does. The economics of hyperscale and colocation shift accordingly, with operators chasing sites that can deliver power at scale.

GPUs, AI factories and the jump in compute

The component at the centre of all this is the GPU. Originally designed to render graphics, GPUs turned out to be ideal for AI because they perform thousands of calculations in parallel, which is exactly what training a neural network demands. A modern AI data centre wires thousands of these chips together so they behave as one large machine.

That assembly, a tightly networked cluster of GPUs working as a unit, is what NVIDIA and the wider industry now describe as an AI factory: a cluster of GPUs run as a single engine for producing AI. It is the reason a new tier of providers, the neocloud operators, has emerged alongside the established hyperscalers, renting GPU capacity by the cluster.

Power and water: what the AI build means for Australian resources

Because AI data centres consume so much electricity, their growth is now a grid question. Operators increasingly secure their own supply through long-term renewable power agreements with solar and wind farms, and pair them with battery storage to smooth demand. Power availability is the constraint that decides which projects proceed and when.

Water draws attention too, because some cooling systems use it to reject heat. The national footprint is smaller than the debate suggests: Australian data centres account for roughly 0.04% of national water consumption, on figures the CSIRO drew from the Australian Bureau of Statistics Water Account. Newer facilities increasingly use recycled water or air-based and closed-loop cooling to keep even that figure down.

How many AI data centres Australia has, and who certifies them

Australia’s data centre map spans 208 tracked sites, from hyperscale campuses in Western Sydney and Melbourne to regional and edge facilities. A growing share are being built or retrofitted for AI workloads, with the largest concentrations of power-dense capacity around Sydney and Melbourne. The full, searchable list lives in our directory of data centres in Australia.

For buyers in government, defence and regulated industries, one further layer matters: certification. A subset of Australian facilities are government-approved and certified under the Hosting Certification Framework, the standard that determines which providers can host sensitive government workloads. For sovereign and regulated AI procurement, certification status is the deciding factor.