AI is drafting emails, generating images and answering questions for millions of us. The data centres that make it possible are going up across Australia, and part of what it costs to plug them in could end up on your power bill.

I spend my working life thinking about what goes into a home and what it costs to run. So here's the no BS version of what's going on, why it matters to households, and what we should ask for.


What a data centre actually is

Picture a shed the size of several football fields, packed with computers running 24/7. They store your photos, stream your shows and now train and run AI models.

All those computers make a lot of heat, so they need industrial-scale cooling running constantly: fans, pumps and chillers. A single campus can draw tens to hundreds of megawatts, around the clock.

AI has pushed this into overdrive. The International Energy Agency expects electricity demand from data centres to surge this decade, and its 2024 analysis said global data-centre use could double by 2026. In Australia, the market operator AEMO now models data-centre growth on its own, with around 5 TWh a year by 2033–34 in its central case and around 15 TWh in an accelerated case.

How it ends up on your bill

When a big new load connects, the local network often needs upgrading: new transformers, heavier cables, sometimes a whole new substation.

  • The data centre pays for its own connection. Think of it as the driveway from the street to its front door.
  • Shared upgrades are spread across everyone. Upgrades to substations and lines that serve the wider area are usually recovered through regulated network tariffs. Distributors charge those to retailers, and retailers pass them through to customers.

You won't see a line on your bill that says "data centre upgrade". You'll just see network charges creep up. Unless the rules change, households and small businesses help pay for infrastructure that serves some of the biggest companies in the world.


The water problem

Power isn't the only issue. Many data centres use evaporative cooling towers, which work like a wet towel over a fan. They're effective, and they use a lot of drinking-quality water.

In Melbourne's west, 19 data centre proposals were seeking around 20 gigalitres of water a year. That's roughly what 330,000 Melburnians use annually. In a country that regularly asks households to cut back during drought, that should give us pause.

Air cooling avoids the towers, but it isn't water-free. The power stations supplying the electricity use water too.

Other places are already pushing back

  • Virginia, US: the world's biggest data-centre market. The local utility proposed a separate rate class for very large, always-on users of 25 MW or more, with long-term commitments and full payment for the equipment they need, so everyone else isn't subsidising them.
  • Ireland: data centres used 21% of the country's metered electricity in 2023. New connections near Dublin now face much tighter conditions.
  • Georgia, US: one utility saw its pipeline of large projects jump from 12 GW to 36.5 GW, forcing an expensive scramble for new generation and network.

If we don't set the rules early, the scale of this industry will set them for us.


What about the jobs?

Construction brings hundreds of workers for the length of the build. Once a site is running, the permanent workforce is small, typically dozens to the low hundreds at very large sites.

That's worth having, but it's not the bonanza the announcements suggest. Compare it with a national program to upgrade existing homes with insulation, solar and efficient appliances. That would keep electricians, plumbers and installers busy in every community for years, and it would cut bills rather than add to them.


What Australia should borrow

  1. Make very large users pay their full cost. A separate tariff class for big, always-on loads, like Virginia's proposal, is the most direct way to stop households subsidising them.
  2. Require peak-neutral connections. New data centres should show they won't add to evening peak demand, using on-site batteries or agreeing to throttle back when the grid is tight.
  3. Recycled water first. Where recycled water networks exist, connect to them by default. In water-stressed areas, require air cooling or closed-loop systems.
  4. Publish the numbers. Quarterly public reporting of energy and water use. Two useful measures are PUE (total site power divided by computer power) and WUE (water used per unit of computer energy).

None of this stops data centres being built. It just makes sure they play by fair rules.


What you can do

  • Make a submission. When a data centre proposal goes on public notification, anyone can make a submission. Ask three things: Will it use recycled water for cooling? Will it be peak-neutral? Who's paying for the local network upgrades?
  • Check your bill. Ask your retailer what share of your plan is network charges, and watch how it changes each year.
  • Talk to your MP. Ask what they're doing to stop households subsidising data-centre infrastructure.
  • If you run a business, look at demand response programs that pay you to cut use at peak times.

The short version

AI isn't going away, and neither should your right to fair bills, secure water and a grid that works. Make very large users pay the full cost they create. Make them peak-neutral. Stop using drinking water when recycled water is available. And make them publish the numbers.


Similar Topics

Cambridge Analytica: How Your Data Became a Weapon
Cambridge Analytica wasn’t just a scandal, it was a preview of how data could be weaponised. In this first post of our six-part series, we break down how they pulled it off—and why the same tactics still shape what you think, how you vote, and what you see online.
CA’s Legacy: Emerdata, Auspex & Data Propria
Cambridge Analytica collapsed in 2018, but its methods never died. Former staff and investors re-emerged in new firms—Emerdata, Auspex, and Data Propria—still using psychological profiling and micro-targeting to influence politics. Same game, same players, new names - your the target!
AI Profiling Explained: Emotion, Movement & Mood Tracking
In 2025, AI is profiling you as you speak, move—even as you sit there. No surveys required. From your smartwatch to your smart car, everything’s feeding invisible algorithms that build emotional profiles on the fly. And yes—it’s creepy, relentless, and a wake-up call to take your rights back.
You Are the Product: AI, Data and Profit
In earlier posts we looked at your privacy, bias, and algorithmic nudges. Now, we discuss surveillance capitalism—a system where your behaviour, emotions, even attention become commodities—and reveal how AI deepens this invisible exploitation.
Emotional AI & Teen Harm: The Cost of Personalisation
AI companions aren’t just chatty — they’re emotionally responsive. That’s helpful, until deep tailoring starts steering how we feel and act. Here’s what the cases, the regulators, and the research say — and how to build (and use) safer tools.
Digital Privacy in the Age of AI: What to Do Now
Think privacy is a lost cause? Think again. We break down data profiling, new laws, and give you a clear action plan—for you, business, and government. Includes a ready-to-use Digital Bill of Rights.