Should Australia develop its own AI, or are we better off surrendering our AI services to be developed by American Tech Bros?

By Dr Martha Knox Haly  

On 11th September 2025, it was reported that Treasurer Jim Chalmers would be meeting with OpenAI executives to discuss the potential for a data centre or cloud computing agreement. A whole-of-government AI service was alreadybeing quietly trialled to accelerate the adoption of AI across the public service. The project began in January 2025, with a Microsoft Based AI hosting service built within GovTeams, which is the Commonwealth secure collaboration platform. The pilot is known as GovAI, and will include models such as ChatGPT4, Google Gemini, and Anthropic’s Claude before November. The first government agency had already been onboarded.

At this point, I have rolled my eyes,’ great’, I thought to myself. ‘We allowed ourselves to be exploited by mining multinationals who took our resources for a song. Now we are about to let in the next big exploitative set of multinationals with the American AI Industry. What is wrong with this country, were we kicked in the head at birth, are we stupid?’

The answer to ‘are we stupid’ is no. The 2023 National Artificial Intelligence Centre (NAIC) AI Ecosystem report found that there are 544 Australian companies whose main product is to develop AI services. They are mostly startups, research institutes, industry organisations and public sector agencies, with several larger publicly listed companies. These companies are a subset of 336,000 technology companies in the professional, scientific and technical services industry. That’s a lot of smart people.

So what did these smart people tell a parliamentary enquiry into AI in 2024? There was agreement amongst experts from CSIRO, Australian Academy of Technological Sciences and Engineering, Science & Technology Australia, and multiple academics on the need for sovereign AI. These experts saw this as an opportunity to develop deep capability. Sovereign AI would be important for privacy, national security, and compliance with local governance. There were strong recommendations from Science and Technology Australia against purchasing foreign off the shelf solutions (which would promote dependency on other countries). Pawsey Super Computing Research Centre and the Curtin University Institute for Data Science advised that Australia had a low level of investment in AI, and needed to spend more. The Kingston AI group described Government investment as an opportunity to develop a premium brand of locally developed AI.

The CPSU (Commonwealth Public Service Union) noted ‘As the use of AI increases, we must continue to build and improve relevant expertise within both Parliament and the public sector. Decision-makers will need greater assistance navigating the issues associated with emerging technologies such as AI. The overload of information available and the need for greater science literacy means that science advisorymechanisms that can help our parliamentarians grapple with these challenges are more important than ever. A dedicated technology assessment office, known as the Parliamentary Science and Technology Office should be established, providing a similar non-partisan and independent function to the Parliamentary Library and Parliamentary Budget Office to inform parliamentary debate.’

Professor Nicholas Davis from the University of Technologyoutlined the benefits of Foundational AI models, deepening Australian understanding of how foundational AI modelswork – with potential spillover benefits in maths, data science, text analysis and health sciences. Professor Anton Van den Hengel (University of Adelaide) echoed similar sentiments, but expressed concern about the shifting of offshore profits, as well as the disruption to development of local mining and agricultural industries through use of foreign AI foundational models. CSIRO and Xaana. AI pointed to the monopolistic power of AI companies and how this could curtail Australia’s economic independence. The experts believed in the importance of sovereign locally developed AI, with some preference for the public sector to take the lead on development, whilst others wanted private sector incentivisation.

So why with all this good advice did we end up with Microsoft building the Government platform which is going to be dominated by American AI firms? The Parliamentary enquiry received evidence that the cost of independently developing a sovereign LLM (large language model) in Australia would be significant. In particular, that the training of AI models required High Performance Computing (HPC)facilities with multiple servers housing multiple graphics processing units (GPUs). As an illustration, the committee heard:

• A single Nvidia H100 (the leading industry benchmark GPU) would cost approximately AUD $40 000-$50000.
• The Nvidia DGX server which houses eight H100 GPUs costs approximately AUD $450000.

There was a detailed reference to how much it had takenOpenAI to develop its foundational model. Monash University argued that Australian models could not be competitive against pre-existing large language model. The Tech Council of Australia countered that Australia’s tech industry was already globally competitive in enterprise software, fintech, quantum computing and biotech. All this was before the Chinese model Deepseek was reportedly developed for $5.6 million and launched in January 2025.

The parliamentary committee expressed concerns about costs and how emissions and could be offset by purchasing pre-existing models. Dissenting reports were tabled from the Greens, the Coalition, and David Pocock. For the record, Pocock was the only one whose dissenting report spoke in defence of Australia’s besieged arts community. Dissenting reports were focused on the environmental costs of data centres. There was an optimistic assumption that extra emissions from AI would be offset by Australia’s transition to a renewable grid. Optimism would be given the lie when the Climate Change Authority revised its original advice for the 2035 climate emissions reduction target down from 65% to 75% down to 62%-70% because of growing electricity demand by data centres.

The culture of outsourcing is another factor. Federal governments across the world (and the Australian Federal Government is no exception) are heavily dependent on IT outsourcing. This means critical IT (including AI expertise) resides in the heads of the contractors and consultants who respond to the market. The expertise does not reside in the heads of Ministers and public servants, who are answerable to us. Public Governance expert, Professor Marianna Mazzucato explains how the outsourcing of complex difficult tasks has resulted in a situation, where the public sector does not know ‘who to work with, or how to write the terms of reference.’ Microsoft, OpenAI, Google and Claude were recognised brands, and this must have been reassuring to ministers and heads of department.

Don’t forget the lobbying. On the 16th June 2023, it was announced that Sam Altman of ChatGPT met the then Industry and Science Minister Ed Husic. This followed the government’s announcement of an 8 week enquiry into the regulation of AI. Altman referred to concensus that AI development should not be curtailed, but regulation was needed to minimise unintentionally harmful product releases. On 30th June 2025, OpenAI’s Chief Economist Ronnie Chatterji visited Federal Ministers, telling an irresistible tale of how AI would make the Australian economy $115 billion larger by 2030. Commentators such as Bernard Keane from Crikey were scathing about the veracity of this figure, referring to it as a self-interested productivity scam. For ministers who believe that the ordinary voter’s quality of life is dependent on increasing economic growth, Chatterji’s patter was like psychological crack. Treasurer Jim Chalmers was coaxed into promoting ‘a light touch’ on AI regulation. The fact that quality of life and longevity is driven by well-funded public health, public education, housing, human connection and life purpose, rather than economic growth was sidelined.

All these meetings were before the public knowledge that:

• A subcontractor to the NSW Reconstruction Authority had uploaded 12000 rows of deeply personal data from 3000 flood victims into Chat GPT in March 2025.
• Deloitte charged the Federal Government $440000 for an AI generated report with fabricated references.
• OpenAI was being sued for ChatGPT4 allegedly facilitating the death of a suicidal teenager.
• There were growing fears of an AI tech bubble, as the market realised that the AI Tech Bros were not really producing substantially improved models.

The timing of Altman and Chatterji’s visits were coincidentally just four weeks before Australia’s Productivity Summit in August 2025. OpenAI had launched an AI Blueprint for Australia in San Francisco, and was contemplating opening an Australian subsidiary. In this blueprint, OpenAI argued for generous tax incentives for businesses investing in AI or adopting AI. Now whenever I hear the phrases ‘generous tax incentives’ and ‘light touch regulation’ in proximity to each other, I automatically think ‘social inequality, goodbye safety and transparency.’ All those problems with transfer pricing, offshoring profits, tax minimisation, and disproportionate influence by the private resources sector elite on democracy that we are complaining about with the resources sector, are about to be replicated in the Australian AI tech sector.

The phrase ‘there is no such thing as a free lunch’ is tired and worn, but it is applicable here. In her book, The Big Con, Professor Marianna Mazzucato describes a phenomenon where private sector providers start by initially offering low price options, as a way of currying favour. This becomes a pathway to seeking bigger favours and extraction of higher costs. We are already seeing this strategy with American AI companies with other governments. For example, Athropic, the creators of Claude, offered Claude to the US Federal Government for $1, with an obvious eye to future government contracts. Then there is the conflict of interest element. It is the job of government to govern, and you can’t effectively govern any entity when it has its fingers in your entrails via GovAI. I have already alluded to the way in which the data centre lobby has reduced the Australian Government’s emissions reduction targets.

Reliance on American multinational tech companies could actually be much worse that what we had already experienced with the resources industry, because data centres are such voracious consumers of water and electricity. We only need to look at Ireland to see what can happen. Ireland hosts 135 data centres. Between 2015-2021, electricity demand by data centres grew by a whopping 265%. By 2030, it has been estimated that data centres will consume 30% of the electricity delivered through Ireland’s grid. Whilst data centres generated job opportunities in the construction phase, a considerably smaller permanent workforce is requiredduring the operational phase. In 2022, Irish data centres consumed 608000 megaliters of water. Data centres are currently competing for electricity and water against hospitals and new social housing in Ireland. The Government has issued a temporary moratorium on new data centres and is re-evaluating its dependence on American IT companies. Ireland is not the only country struggling with this; a recent BBC story reported that Scottish data centres were using  the equivalent of 27 million bottles of water each year.

Data centres have a capacity to throw a spanner in the works as Australia is trying to head to net zero emissions. It is has been estimated that CO2 from software related activity and data centres will rise to 14% of all emissions in 2040. Now consider that Australia is a hot, dry country, where electricity for cooling in summer is essential for staying alive in many parts of the country. We already have communities in this country that do not have adequate water supply. We also know that a renewably powered grid can’t possibly meet emissions targets, when continual economic growth keeps lifting the bar. Australia currently hosts 273 data centres. So the dystopian prospect of balancing resources which are essential for human survival against the needs of privately owned AI Data Centres is a real possibility.

Timing is another significant aspect to consider. The Parliamentary enquiry, the Productivity Summit and Chalmer’s September 23rd 2025 announcement happened before the US Senate passed the GAIN AI bill on 10th October 2025. The GAIN AI bill mandates prioritisation of NVIDIA chips to American companies over exports, and the House will either pass or amend this bill. So there is a possibility that Australia may not may not even get the chips that are needed to operate GovAi. Just as Nixon burned American allies blowing up the global monetary system on 15th August 1971, right now the Trump administration is burning allies by blowing up global trade.

The Trump administration is breaking freedom of expression, trying to impose political censorship and defunding climate change science initiatives. The administration is actively coercing American traditional media and social media companies, so of course they are going to coerce AI companies. Reliance on American AI firms means that Australia is vulnerable to the critical data pipeline being closed off. When Elon Musk tampered with Grok, we saw how AI can be made to lie at the whim of a single individual. Australians have to stop being naïve about what this means for us as a country when our Government is intimately connected with American AI platforms. If we stay connected, and assuming we are actually getting the chips needed to run GovAI, we will be part of the Trumpian climate denialist twilight zone. If Australia doesn’t get the Nvidia chips, then we have to to run our government AI platform within an entirely sovereign Australian AI ecosystem (despite Professor Nicholson’s injunction that we avoid ‘moonshots’ in developing Australian AI).

Is Australian Sovereign AI possible, and can this be done locally in Australia? Right now sovereign AI does not appear to be possible in any middle power country, but severalmiddle power countries which have taken on the goal ofworking towards sovereign AI. For Australia, it would be hard and costly. Australia would certainly be starting from a lower base than say South Korea, which is building its own AI infrastructure. Whilst the Korean consortium trains its models on AI Nvidia GPUs and develops data centres with Amazon Webservices, South Korea is currently heavily investing to reduce those dependencies. Switzerland has recently launched the Apertus LLM which was built by Swiss Universities in 2025. Apertus is a no frills open source model and blue print for a safer, more accessible AI LLM for scientific researchers and commerce. The open source statusmeans the design manual and source code is available for public scrutiny, and  a copy can be downloaded to user’s own servers protecting user’s data privacy.

The challenge of things being difficult, hard and costly is not a reason for avoiding AI sovereignty. Malaysia is a poorer country than Australia, and found itself to be the rag doll in a potential tug of war between the US and China in the AI sphere. Malaysia choose to opt out of using either Huawei or Nvidia AI chips, by manufacturing its own. The chip is less effective than NVIDIA, but Malaysia sensibly removed itself from a geopolitical tug of war. If Malaysia can do it, so can Australia.

A five year pathway with steps and rough costings has been outlined by the Australian Academy of Technical Sciences and Engineering (ATSE). ATSE proposed a $5 billion plan for:

• The establishment of 5-7 Government owned AI factories.
• An AI Supercomputing hub (initially 5,000–10,000 GPUs).
• Establishment of an AI Safety Institute.
• As well as the creation of a  National Trusted Data Lake (multi-sector)  and secure cloud for critical infrastructure.
• Foundation model training and 41 domain-specific sovereign models in areas of national priority.
• AI testbeds and regulatory sandboxes – all run out of the public sector.
• ATSE also proposed MOUs with Singapore, Korea, EU and UK (but very tellingly not the US).

ATSE recognised that public ownership facillitates regulation of data centres, and ensures that research centres (as opposed to purchases on Amazon web) would be given priority. This public sector digital infrastructure would be integrated with the National Computational Infrastructure, the Pawsey Centre, and state data centres.

ATSE identified current gaps in  a scalable public AI infrastructure. The most pressing gaps were:

• Reliance on OpenAI, AWS and Google.
• Lack of a dedicated computing infrastructure for training foundation models or domain-specific models.
• No national-scale optimised GPU cluster accessible to public researchers and industry (e.g. 10,000+ GPU-class system).

These gaps meant there was no flexible allocation for AI research workloads, resulting in queue times which were a problem for iterative AI development. Absence of flexible allocation capacity meant that the Australian Energy Market Operator is running the grid at full capacity, with an increased risk of blackouts. ATSE explained that dependence on multinational tech companies was deepened by the lack of large-scale domain specific AI models due to limited digital infrastructure. $5 billion sounds like a great deal of money for Australia to spend on AI infrastructure, but consider Taiwan (a smaller country) is investing $10.53 billion for just the next three years.

AI is embedded in the Australian landscape. Privately owned data centres are causing emissions to rise, suctioning up electricity and water, and the data centre lobby is already distorting climate change policy. The number of private data centres needs to be capped and the number of publicly owned data centres needs to be kept to the small numbers outlined in ATSE’s report (and no more). Public ownership offers the best chance of data sovereignty, data privacy, and regulatory control in the interests of Australians and the environment. This is only possible if we reduce independence on American AI multinationals and embrace the hard ambitious task of creating a fully sovereign Australian AI digital infrastructure. It means starting building publicly owned AI factories for Australian AI chip manufacturing, developing Australian foundational AI models, and prioritising our researchers and creatives over AI slop. Within this framework, we need to ask why do we need AI? AI is unquestionably valuable for physical sciences and health research, but should this privilege be extended to creating images, texts, writing better emails, or crunching financial risk for home loans? Is writing an email quickly or creating quirky cat images really worth sacrificing our water and climate, whilst displacing huge numbers of Australians into unemployment? I will leave you to think about this, and if you are concerned, call your local MP.

Keep Independent Journalism Alive – Support The AIMN

Dear Reader,

Since 2013, The Australian Independent Media Network has been a fearless voice for truth, giving public interest journalists a platform to hold power to account. From expert analysis on national and global events to uncovering issues that matter to you, we’re here because of your support.

Running an independent site isn’t cheap, and rising costs mean we need you now more than ever. Your donation – big or small – keeps our servers humming, our writers digging, and our stories free for all.

Join our community of truth-seekers. Donate via PayPal or credit card via the button below, or bank transfer [BSB: 062500; A/c no: 10495969] and help us keep shining a light.

With gratitude, The AIMN Team