Anthony Albanese stood before a crowd at the University of Sydney and drew a line in the sand. He promised that Australia will not just be a passive spectator in the technological shift of our era. He told us we will not become a mere "data warehouse for AI products made overseas."
It is a beautiful sentiment. It is exactly what you want to hear from a prime minister.
The government plans to build a brand-new national Office of AI, force data centres to pay for their own electricity generation, and shield local artists from having their work scraped by international algorithms. Albanese even called unauthorized scraping "theft."
But let's be honest. We have seen this movie before, and the ending is usually written in Silicon Valley, not Canberra.
The grand strategy sounds admirable, but it ignores a glaring truth. The tech empires building these models have more capital, more compute power, and more global leverage than most sovereign nations. If Australia pushes too hard with rules that are impossible to enforce, these companies will not bend. They will simply bypass us.
The Core Promises of the New Blueprint
The government's proposal tries to solve three massive, interconnected issues: energy, intellectual property, and regulatory fragmentation.
First, there is the immediate problem of physical infrastructure. AI requires an unimaginable amount of power and water. Data centres are popping up across the country, causing friction in local communities over land use and soaring utility demands. The new policy says operators must underwrite their own new renewable energy generation. They will have to pay the full cost of grid connections so regular families do not see their power bills climb.
Second, the plan takes a firm stand on copyright. Tech companies have been lobbying for years for "text and data mining" exemptions. They basically want permission to scrape the entire Australian internet for free to train their models. Albanese explicitly rejected this. He declared that no company should use Australian books, music, or news without permission and payment.
Third, the government is creating a central regulator: the Office of AI, housed directly inside the Department of the Prime Minister and Cabinet. The goal is to speed up approvals while holding developers to strict national standards.
It all sounds incredibly neat. But when you look at the mechanics, the plan starts to show some major cracks.
Why Big Tech Does Not Fear Canberra
Australia has a history of trying to bully tech giants. Sometimes it works. Often it backfires.
Think back to the News Media Bargaining Code. Australia tried to force Facebook and Google to pay local publishers for news links. Facebook responded by turning off the news feed for the entire country overnight. Emergency services, health departments, and community groups were suddenly blocked from posting. It was a stark reminder of who actually holds the keys to the modern public square.
The stakes with AI are infinitely higher. If Google or Meta decides that Australia's local copyright laws are too restrictive, they will not negotiate a complex licensing deal for a market of 26 million people. They will simply block their advanced models from being IP-resolved in Australia.
We are already seeing this happen elsewhere. Apple delayed the rollout of its intelligence features in the European Union because of regulatory disputes. Meta did the same with its multimodal models. If the EU, with its market of hundreds of millions of consumers, struggles to dictate terms to these companies, Australia has very little leverage.
If we get cut off from the most advanced tools, our local businesses, universities, and developers will fall behind. That is the leverage these corporations hold. They know we need them more than they need us.
The Grid Fantasy and the Data Centre Surge
Let's look at the energy rules. The plan says large data centres must underwrite new electricity generation to keep consumer bills down.
On paper, this is brilliant. In reality, it is a logistical nightmare.
Data centres run twenty-four hours a day, seven days a week. They need constant, baseline power. You cannot run a global cloud system purely on solar power that only works when the sun shines, or wind power that dies down at night, unless you have massive, expensive battery storage.
If a tech company wants to build a new data centre in Western Sydney or Melbourne, they cannot just build a wind farm next door and call it a day. They have to hook into the national electricity grid. If the grid is already strained, adding a facility that consumes as much electricity as a small city will inevitably push up baseline prices, no matter how many green offsets the company buys.
If we make the energy requirements too hostile, those investments will simply walk away. The Australian Industry Group has already warned that over-regulation could drive billions of dollars of data centre investment to other countries in our region. We cannot build a domestic tech industry if we do not have the physical computers to run it.
The Impossible Copyright Policing Mission
Albanese's defense of Australian creatives is noble. He said using our artists' work without their control is theft. The crowd cheered.
But how do you actually police this?
An AI model is not a copy-paste machine. It does not store a database of Australian books and music inside its code. It digests them, extracts the patterns, and discards the original file. Once a model is trained, it is incredibly difficult to prove that a specific Australian novel was part of the training data unless the model spits out a direct plagiarism of the text.
Even if we pass laws saying companies cannot scrape Australian websites, those companies are not based here. They train their models on servers located in Virginia, Dublin, or Singapore. Australian copyright law stops at our borders. We cannot easily sue a US-registered company in a US court for scraping a public website using Australian intellectual property laws, especially when US courts are still trying to figure out their own fair use rules.
If we ban local scraping, we might just end up stopping Australian AI startups from training on local data, while the global giants continue to scrape whatever they want from overseas servers. We risk hobbling our own innovators while doing nothing to stop the global giants.
The Swiss Cheese Dilemma
Even within the Labor party, there is skepticism about this slow, methodical approach.
Ed Husic, the former industry minister, did not hold back his criticism. Before being moved out of his cabinet role, Husic wanted a sweeping, comprehensive legislative act. He warned that the current plan looks like "Swiss cheese" and that the government should not just deliver a "fancier cheeseboard." He pointed out that the government has responded faster to food-tampering scares than to the systemic risks of unregulated AI.
He is right. A voluntary standard here and a planning guideline there will not stop deepfakes from ruining lives, algorithms from driving discrimination, or automated systems from wiping out entry-level jobs.
By pushing the actual legislation to early next year, the government is trying to buy time. They want to consult, hold national cabinet meetings, and draft perfect rules. But while we draft papers, the technology is evolving weekly. By the time this law passes in 2027, the models we are trying to regulate today will be obsolete.
Actionable Steps for Australia
We cannot stop the tide, but we can stop pretending we can control it with press releases. If Australia wants to protect its sovereignty and its people, we need a pragmatic strategy, not just an admirable plan.
Here is what we should actually do:
- Fund localized compute instead of just regulating it. If we want sovereign AI, we need sovereign hardware. The government should directly co-fund public-private partnership data centres dedicated solely to research, local business, and government services, bypassing the big tech gatekeepers.
- Focus on practical liability, not scraping bans. Instead of trying to police what goes into the black box of an AI model overseas, focus on the output. If an AI system deployed in Australia produces defamatory content, breaches privacy, or facilitates fraud, hold the company deploying it strictly liable under existing consumer protection laws.
- Build a national licensing clearinghouse. Make it easy for tech companies to do the right thing. Create a centralized, government-backed collective licensing body—similar to APRA AMCOS for music—where creators can opt-in to license their data for AI training at standardized, fair rates.
- Mandate water and energy efficiency targets, not just offsets. Do not let data centre operators buy their way out of local environmental damage with cheap carbon credits. Force them to use direct liquid cooling and build onsite renewable storage that contributes to local grid stability during peak hours.
We cannot shame Silicon Valley into playing by our rules. We have to build our own leverage, accept the physical realities of the grid, and protect our artists with laws that can actually be enforced in the real world. Anything else is just wishful thinking.
.jpg)
