If I built a Data Centre

A data centre is not automatically good for humanity. It is a huge secure 'shed' that converts electricity, water, land and computer equipment into digital processing.

It becomes valuable when it supports medical research, emergency communications, weather forecasting, agriculture, banking, education, government services and Australian businesses. It becomes dangerous when it is mainly used for surveillance, behavioural manipulation, advertising extraction, financial speculation or concentrating control of information in a handful of companies.

The building is neutral. Ownership, purpose and control determine whether it serves people or farms.

What building one actually involves:

A proper data centre is simultaneously:

• A major electricity consumer with its own substations, transformers, switchgear, batteries and generators.
• A cooling plant operating every minute of the year.
• A secure industrial building.
• A telecommunications exchange connected through multiple independent fibre routes.
• A cybersecurity and critical-infrastructure operation.
• A property, energy and long-term leasing business.

The first question is therefore not, “Where do we buy land?” It is:

Who will contract to use the computing capacity and pay for it for the next 10 to 20 years?

Building one speculatively without an anchor customer would be financial suicide.

Rough Australian cost:

Published Australian market estimates currently sit around A$9 million to A$14 million for every commissioned megawatt, before allowing for some land purchases, difficult grid upgrades and the actual computer equipment. Australia is among the most expensive data-centre construction markets in Asia-Pacific, while AI-ready facilities can carry a further construction premium because of their higher electrical density and liquid-cooling requirements.

 

Those are feasibility-level numbers, not quotes. AI chips and networking can cost considerably more than the building holding them.

You would also need development capital before construction: land options, engineering, grid studies, planning, environmental work, customer negotiations, legal structure and financing. That preparatory phase alone can consume several million dollars.

Site required

Land is important, but available electrical capacity and fibre are much harder to find.

A sensible guide for this would be:

• 0.5–2 MW: an existing industrial property or roughly 0.5–2 hectares.
• 5 MW modular campus: approximately 2–5 hectares.
• 20 MW campus: approximately 5–15 hectares.
• 100 MW hyperscale development: perhaps 15–50 hectares, depending on density, generation, cooling and future expansion.

Compact urban facilities can fit on much smaller sites. For example, AirTrunk’s previously announced 110 MW Sydney facility occupied about four hectares. That illustrates why there is no reliable hectares-per-megawatt rule.

The site needs:

Power: Two independent grid supplies if possible, or a grid connection backed by on-site generation and batteries. A nominal 10 MW computing load will typically require something closer to 12–15 MW continuously once cooling and building systems are included.

Fibre: At least two, preferably three physically separate fibre routes entering from different directions. Two carriers using the same underground trench are not genuine redundancy.

Cooling: Air cooling for ordinary workloads; direct-to-chip liquid cooling for dense AI equipment. Water availability, wastewater, heat rejection, noise and local climate all matter.

Physical protection: Away from floods, bushfires, unstable ground, hazardous neighbours and aircraft risks. Cyclone exposure changes the construction standard considerably.

Approvals: Planning, electricity connection, fuel storage, noise, water, emissions, building, environmental and critical-infrastructure obligations.

Operations: Twenty-four-hour engineering, security, network monitoring, cybersecurity, maintenance, incident response and customer support.

The grid connection may become the deal-killer. Australian data centres generally have to fund their direct connection and potentially contribute toward upstream network upgrades they bring forward. Locating in a congested part of the network can substantially increase the capital contribution.

Australia versus India

Australia is not universally “the best place.” It depends on what the centre is meant to do.

Australia wins when the product is trust and sovereignty

Australia is stronger for:

• Australian government, defence, health, financial and critical-infrastructure information.
• Customers who need Australian law, courts, auditing and physical access.
• Businesses wanting lower geopolitical exposure.
• Regional disaster recovery and national continuity.
• Australian research, agriculture, health and AI development.
• Customers prepared to pay more for demonstrable control.

Australia also has political stability, large areas of land, strong engineering capability and enormous potential renewable-energy resources.

But Australia has serious disadvantages: expensive construction, expensive skilled labour, grid congestion, slow connections, limited transformer and electrical-equipment supply, community resistance and a relatively small domestic population.

AEMO’s work shows the scale of the energy problem. Australian data centres were estimated to use 3.9 TWh in FY2025, around 2% of National Electricity Market consumption. Under its central scenario, that rises to roughly 12 TWh and 6% of NEM electricity by 2030. A data-centre boom without additional generation and transmission could compete with homes, farms and ordinary industry for power.

India wins when cost and market size dominate

India has much cheaper land, construction and labour, an enormous domestic customer base, major technology workforces and strong government support for data-centre investment. One recent market report estimates Indian development costs at approximately US$5.5 million to US$8.5 million per megawatt, among the lowest globally. India’s capacity is also expanding rapidly because its market remains under-supplied relative to its number of internet users.

For Indian banking, telecommunications, commerce and consumers, India makes perfect sense.

For Australian sovereign information, India does not.

The additional network distance, foreign jurisdiction, offshore privileged access and dependence on another country’s legal and political system defeat much of the sovereignty argument.

On Indian companies holding master administration

Outsourcing privileged administration overseas is a legitimate concern. 

The real problem is this:

No outside company, Australian or foreign, should possess unrestricted standing master access to sensitive Australian systems.

An Australian administrator can steal information. An Indian administrator can steal information. An American cloud company can be compelled by American law to provide information. A foreign shareholder can influence security decisions. The threat is uncontrolled privilege, conflicting laws and lack of transparency, not ethnicity.

Australia’s Critical Infrastructure Security Centre specifically warns that foreign-owned or operated facilities may be compelled under foreign law to provide access to information, potentially without the customer’s knowledge, and that some foreign laws can extend beyond the country where the physical equipment sits.

A genuinely sovereign Australian data centre would therefore require:

• No permanent offshore administrator accounts.
• Australian control of the cloud management plane, not merely Australian storage.
• Customer-owned encryption keys held in Australia.
• Just-in-time administrator access that automatically expires.
• Two-person approval for sensitive access.
• Full administrator-session recording and independent auditing.
• Separation between the facility owner, system operator and encryption-key custodian.
• Australian incident-response and disaster-recovery authority.
• A contractual ability to disconnect any overseas vendor immediately.
• No hidden subcontracting of privileged functions.

Simply putting servers in Brisbane while an overseas corporation controls the software, encryption keys, updates and administrator accounts is data residency theatre, not sovereignty.

Where Australia could genuinely outperform India.

Australia should not try to beat India by offering cheaper labour. We will lose.

Australia could win by offering something more valuable:

A politically stable, independently powered, Australian-controlled computing refuge for the Asia-Pacific region.

The strongest model would combine:

• Dedicated new electricity generation rather than draining the existing household grid.
• Large batteries and perhaps dispatchable backup generation.
• A campus near multiple international and national fibre routes.
• Australian ownership of the operating entity and control systems.
• Transparent water and energy reporting.
• A second geographically separate Australian recovery site.
• Computer reserved for public-interest work such as weather, fire, flood, agriculture, health and scientific research.
• Training and skilled employment for regional Australians.
• Strict prohibition on selling customer data or using it to train unrelated systems.

Darwin and Perth have strategic proximity to Asia but harsher cooling and power challenges. Tasmania and southern Victoria have cooler climates and renewable resources but require strong cable and power redundancy. The Canberra–Sydney corridor is attractive for sovereign government work but expensive and increasingly power-constrained. South Australia has energy and land potential but would need first-class fibre and anchor customers.

The right balance could be two medium-sized facilities rather than one giant one: one close to eastern Australian customers and one on a different grid and telecommunications route for genuine national resilience.

What I would build

I would not start with a 100 MW AI monument.

I would examine a 2–5 MW modular Australian Sovereign Resilience Centre, expandable in stages:

• Australian-controlled privileged access.
• Dedicated power generation and storage.
• Two genuinely independent fibre paths.
• Liquid-cooling capability for selected AI racks.
• Customers in regional health, councils, agriculture, emergency management, financial services, NDIS systems, universities and Australian small businesses.
• A second recovery location planned from day one.
• A protected allocation of computing capacity for food security, weather, disaster response and regional research.

That is commercially possible, nationally useful and differentiated. Another generic shed leasing racks to multinational cloud companies would simply hand them another piece of infrastructure while Australians provide the land, electricity and water.

The opportunity is not merely storing Australian data in Australia. It is keeping the authority to operate, encrypt, inspect, recover and ultimately switch off that system in Australian hands.

And from my position I see a  farmer’s land as being boxed in by acres of solar panels to feed a data centre, while the farm loses its space, its future and eventually its right to exist. Losing land rights.... is another topic.

 

Bronwyn Holm, Founder, Earthfood®
Farmers’ Friend • Gardeners’ Guide • Soil Advocate • Growers’ Voice

© 2026 Bronwyn Holm. All rights reserved. Earthfood® • Earthfood Pantry™ • Earthfood Conversations™

Shop living soil microbes at yourearthfood.com.au

Earthfood® certification and standards portfolio: USDA Certified • OMRI Listed • Southern Cross (AU) Certified • Australia Organics Certified • Ecocert (EU) • Organic Trade Association Approved • Carbon8 and Regener8 Standards • International Vegan Association Approved • International Harmony Code.

AI and data-mining rights reserved. This work may not be scraped, copied, ingested, reproduced, adapted or used to train any artificial intelligence or machine-learning system without prior written permission.