Joint Statement on Sovereign AI Capability in Australia

Australian Council of Trade Unions (ACTU)

University of New South Wales (UNSW) AI Institute

Adelaide University Australian Institute for Machine Learning (AIML)

Kingston AI Group (KAIG)

The time is now to determine the future trajectory of Artificial Intelligence (AI)

In this unique joint statement, the union movement and Australia’s leading AI researchers have come together to propose a government-coordinated national endeavour to enhance Australia’s Sovereign AI Capability. This will be delivered through investments in safe and responsible Australian-led AI development that will boost national economic priorities to create safe, secure, and well-paid jobs. Comparable countries around the world are already investing big to secure their sovereignty: Singapore, Germany, Sweden, Denmark, Japan, Canada, Korea and the European Union among others.

Sovereign AI Capability means more than where AI systems are physically located. It means that Australia can build, adapt, evaluate, operate and sustain AI systems on compute and data infrastructure that Australians own or control, including when access to foreign services is disrupted. It also means ensuring that AI used in Australia is accountable to the Australian people, operates according to Australian laws and social expectations, strengthens our national capability, and generates economic value that is retained in Australia and shared fairly with working people and across the community.

We strongly endorse the Prime Minister’s urgent warning that Australia must be ‘much more than a data warehouse for AI products made overseas.’ This joint statement outlines a specific proposal on how we can build on existing Government commitments and take the next logical step in enhancing Sovereign AI Capability in a way that is aligned with Australia’s economic priorities and social contract. At the heart of this is a national initiative to prioritise the development of AI that augments work and supports working people to do the skilled work that they do. We reject an economic future in which AI and automation are used simply to displace labour, degrade wages and job conditions, the theft of creative workers’ outputs, and value captured largely by companies located offshore. Used well, AI and robotics make work safer and more productive and keep firms and jobs in Australia that would otherwise move overseas.

To achieve this, we propose a government-coordinated effort bringing together unions, universities, industry experts, the superannuation sector and the Australian technology industry. It would strengthen Australian control over AI infrastructure and support the development of a portfolio of Australian AI models, from domain-specific models to Australian foundation models, whose use in the workplace is co-designed with workers and unions, that meet the needs of our economy and support the creation of secure, safe and high-quality jobs.

This will be delivered through the creation of a Sovereign AI Coordination Vehicle by the Australian Government. This proposal is outlined in detail below.

The plan to build sovereign AI capability in Australia

The argument in favour of substantive investment in Australia’s own AI industry to boost national economic priorities and support the creation of secure and safe jobs is overwhelming, as is the necessity for Sovereign AI capability for national and cyber security in increasingly uncertain global geopolitical and economic environments.

Proposal

To realise our common ambition of developing the AI industry to support national economic priorities and the creation of safe, secure, and well-paid jobs, we are jointly calling for a Sovereign AI Coordination Vehicle to be established by the Australian Government. This will represent a government-coordinated effort bringing together unions, universities and industry experts, the superannuation sector, and the Australian tech industry to enhance our control over AI infrastructure and build our own AI models, including Australian foundation models, deployed with worker consultation and suited to the needs of our economy creating secure, safe, highquality jobs for the future.

The coordination vehicle

A. Will be a standing tripartite body established with foundational corporate members, union representatives, and the Australian Government, with the research sector represented on the board through universities and publicly funded research organisations, to undertake the fundamental research and commercialisation support required to create global competitive advantage for the Australian AI industry while supporting national economic priorities and creating safe, secure, and well-paid jobs. This is critical if our companies are to develop and pursue global market opportunities.

B. Will be established as a not-for-profit company with a tripartite board, so that it can move with the speed and commercial flexibility required to deliver a competitive advantage for Australian industry, while ensuring that intellectual property generated with public funding or public data remains in Australia and returns value to the Australian public. It could take the form of, or house, an AI Research and Development Corporation modelled on the rural research and development corporations.

C. Will invest in Australian-controlled compute, data infrastructure, secure development and testing environments, engineering support, and a portfolio of Australian AI models and applications. Compute would be commissioned rather than built from scratch, operated by Australia’s existing national research infrastructure providers and Australian-owned neocloud providers, and allocated through the established national merit process, so that capacity forms a single national pool rather than fragmenting across institutions, states and vendors. New corporate partners can join, including investment funds, based on ongoing financial commitment, alignment with values and mission, and acceptance that the Vehicle’s compute, data and intellectual property remain under Australian control.

D. Will invest in and assist in coordinating training across the full AI capability chain: AI researchers and builders, research software and infrastructure engineers, technicians, data stewards, security specialists, AI assurance evaluators, domain experts, tradespeople and workers who will use AI. It will bring world-class AI skills to the Australian people. Training will be vendor-neutral, building transferable capability rather than familiarity with particular products.

E. Will be funded through Commonwealth investment alongside contributions from the AI infrastructure sector. Data centre and AI infrastructure developers seeking Commonwealth or state facilitation should be required to reserve a defined share of installed AI compute for the national pool, discharged in kind or by trade to another provider, and to make an AI Capability Contribution that funds the people, software and data infrastructure needed to use it. These obligations should bind the party that controls the compute, with contractual flow-down from data centre operators to their tenants. Free tokens, expiring credits and vendor-specific training should not count toward them. Major AI companies doing substantial business in Australia should also maintain a genuine research and development presence onshore.

How the coordination vehicle would develop Australia’s sovereign AI capability

The coordination vehicle would oversee the development of Australian AI models, from domain-specific to foundation models, and seed the development of a portfolio of missiondriven projects in areas of Australian need and comparative advantage, including health, energy, climate, critical minerals, agriculture, manufacturing, defence, public services, education and Australian language and culture, with any use of First Nations data governed by Indigenous Data Sovereignty and the CARE principles. Projects should bring together companies, workers, research organisations, universities, TAFEs, investors and government around defined outcomes, including supporting and creating secure, safe, and well-paid jobs. It will invest in the vital AI infrastructure Australia will require to cultivate our own AI industry, not just to support the operations of US-based multinationals in Australia.

Sovereign AI models do not require the multibillion-dollar development funding of the kind received by such AI titans as OpenAI, Google or Anthropic. Australia needs the capability to build a variety of models to suit our economic needs (including but not limited to Australian foundation models), adapt open-weight models, develop models specifically for Australian conditions, evaluate foreign systems independently, and keep critical services operating if access to models is disrupted. Switzerland has built an open-source AI model, trained across more than 1,000 languages, including several that are normally not featured in AI models, such as Swiss German, on infrastructure costing a relatively modest AU$70 million annually. Its architecture, training data, methods and weights are publicly documented, and its governance sits with public institutions. Australia should adopt the same principle: public control where public data and public money are central, openness where safe and appropriate, and restricted deployment where sensitivity requires it.

Specific compute infrastructure is required to develop and train AI models. Australia’s existing data centres and supercomputer facilities lack the scalable, sovereign GPU infrastructure to support domestic AI model development. Australia’s entire institutional GPU base across universities, CSIRO and national facilities is around 2,200 accelerators, most of them several generations old, and public research receives an estimated one to five per cent of usable frontier GPU-hours nationally. Frontier Australian research is already moving offshore for want of domestic capacity. Local data centres primarily focus on providing cloud infrastructure. The few that have GPU hardware are focused on inference for overseas models rather than training Australian ones. This reflects the fact that local inference capability is what foreign model retailers require to rent their services to Australians.

Without such local training capacity, Australia risks becoming a passive consumer in the AI value chain, with limited ability to innovate, customise, or audit these disruptive technologies.

The Coordination Vehicle would invest in vital AI infrastructure, such as the development of sovereign AI compute to support the development of locally customised AI models and provide GPU access for startups through the single national pool described above. Other countries are already investing in domestic AI infrastructure:

• The US has announced joint investments with NVIDIA and AMD to build several new supercomputers with the Department of Energy exceeding 2,000 exaFLOPS capacity (more than 100,000 times bigger compute compared to Australia’s NCI supercomputer).
• Under its £2 billion Compute Roadmap, the United Kingdom has committed to a twentyfold increase in public AI compute by 2030. A single one of its public AI systems, Isambard-AI, has 5,448 current-generation superchips, more than double Australia’s entire institutional GPU base.
• Norway has committed over AU$330 million to build its own supercomputer and fund the development of AI models.
• The Korean government has recently announced the planned acquisition of 260,000 GPUs from NVIDIA. The Ministry of Science and ICT are buying 50,000 of these to develop independent AI foundation models and build a national AI computing centre.

By empowering the Coordination Vehicle to seed and coordinate the development of multiple partnerships and initiatives between universities and the private sector, the government would provide Australia with the best opportunity for success.

Sovereign AI and industry policy should be underpinned by government procurement: one of the most effective economic levers the government has. The Coordination Vehicle could establish regional AI factories: AI hubs that accelerate new AI businesses by connecting data, compute capacity, and talent with universities, businesses, and investors. The EU has announced AI factories as part of a broader InvestAI initiative, supported by an AU$360 billion public–private investment that has attracted Airbus, Volkswagen, and Siemens as major partners. The EU plays a strong role in ensuring the optimisation of resourcing, articulated through its AI Continent Action Plan.

Building on the Buy Australia Plan, government tech procurement should explicitly seek to keep value created in Australia within Australia, and to prioritise the creation of well-paid, secure, and safe jobs throughout the supply chain. South Korea shows what this looks like in practice; its AI for All Project will give every citizen free access to a government-supported AI service, with at least half of each supported service required to use Korean models, giving domestic developers customers, compute and operating experience at national scale.

Australia has the talent, sector expertise and demand to build substantial AI capability. What it lacks is a funded national ambition and coordinated access to compute, skills, research support and patient capital. A Sovereign AI Coordination Vehicle should close these gaps and ensure that AI investment strengthens Australia’s economy, workforce, scientific capability and national security.