AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 16

Per Capita

Submitter group: Research institutes.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Per Capita proposes public equity, revenue apportionment, an access fee, reserved compute and training obligations to capture AI infrastructure value across uncertain futures. It proposes separate creative-sector displacement support without copyright changes. The report does not necessarily reflect roundtable participants’ views and offers one defensible model for comparison, with calibrated mechanisms intended to preserve investment viability.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. Argues that Australia's bargaining leverage over AI infrastructure is temporary and typically shifts to asset owners once infrastructure is in place.

    These unique advantages mean Australia currently holds important bargaining power, however, they are only temporary. Once infrastructure, grid connections, and supply chains are in place, leverage typically shifts to the firms that own and operate the assets, as seen in both international experience and Australia's offshore gas industry.

  2. Argues that no single instrument (tax, access fee or equity holding) is likely to secure the national interest across plausible futures, so a layered approach is needed.

    Its central argument is that no single policy instrument, whether a tax, access fee or equity holding, is likely to secure the national interest across the range of plausible technological and economic outcomes.

  3. Recommends a public equity stake in geographically fixed enabling assets such as data centres, transmission and renewable generation as the foundation of the model.

    The foundation is a public equity stake in the enabling assets themselves, the data centre campuses, compute facilities, transmission infrastructure and the renewable generation behind them.

  4. Recommends reserving a share of sovereign compute in kind for universities, public-interest research, start-ups and the national science agency.

    Reserve a share of sovereign compute in kind for universities, public-interest research, start-ups, the national science agency and strategic needs, as a condition of access to Australian infrastructure.

  5. Argues that creative-sector displacement is distinct from copyright licensing and warrants its own support mechanism funded partly from value-capture proceeds.

    The recommendation is therefore to design a creator-support mechanism for ongoing structural displacement, distinct from and additional to copyright licensing, funded in part from the value-capture proceeds, so that the sector most exposed to AI shares directly in the dividend its work helps generate.

  6. Recommends a local-talent and apprentice commitment, with an apprentice-to-tradesperson ratio tied to grid-connection and planning approvals as one option.

    Build on the legislated skills baseline with a genuine commitment to local talent and apprentices. With training and apprenticeship requirements set to become part of the mandatory data centre standards, the question is how to make them meaningful; an apprentice-to- tradesperson ratio attached to grid-connection and planning approvals is one option worth serious consideration

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Establish a specific economic partnership framework for the AI industry combining public ownership, revenue sharing and access fees, so benefits are shared among all Australians.

    The Australian Government should establish a specific economic partnership framework for the AI industry, combining public ownership, revenue sharing and access fees to ensure any benefits are shared between all Australians.

    A program corrected obvious text-recognition errors in this quote for display.

  2. Adopt a layered value-capture model rather than a single instrument.

    Adopt a layered capture model rather than a single instrument. No one mechanism is both ambitious and robust; the strengths of each offset the weaknesses of the others.

  3. Choose the model for robustness across possible AI futures, not for any single forecast.

    Choose the model for robustness across futures, not for any single forecast. Because AI’s trajectory is deeply uncertain, the test of a good design is that it captures value whether progress stalls or accelerates, not that it is optimised for the future we happen to expect.

  4. Take public equity in fixed data centres, compute, transmission and renewable generation through the Clean Energy Finance Corporation, NBN Co or a new entity, negotiated before commitment.

    Take a public equity stake in the fixed enabling assets (data centre campuses, compute facilities, transmission and the renewable generation behind them) through existing vehicles such as the Clean Energy Finance Corporation, and NBN Co, or a new entity, negotiated before commitment, when stakes are cheapest. Emphasis must be placed on the balance of assets, not staking all investment just in compute.

  5. Balance public investment across compute and inference infrastructure, rather than investing only in compute.

    And again, for emphasis a further reason why there must be balance in investment in inference and compute infrastructure.

  6. Structure investment as genuine public-private partnerships with public co-investment, board representation, transparency and step-in rights, rather than passive grants or tax concessions.

    Structure investment as genuine public-private partnerships with public co-investment, board representation, transparency and step-in rights, not as passive grants or tax concessions.

  7. Give public stakes information rights, transparency over training runs and model deployment, and step-in rights in defined circumstances; condition co-investment on these rights and compliance with government infrastructure expectations.

    A public stake should carry board representation, information rights, transparency over training runs and model deployment, and step-in rights in defined circumstances. Public co-investment should be conditioned on the partner accepting these terms, and on compliance with the Australian Government’s Expectations of data centres and AI infrastructure developers, which already set out principles covering national interest, energy, water, skills and research access.

  8. Apply an ongoing revenue apportionment anchored to countable licence units (registered models, deployment instances) rather than accounting profit, so it cannot be eroded by transfer pricing.

    Apply an ongoing revenue apportionment anchored to countable licence units (registered models, deployment instances) rather than to accounting profit, so that the charge scales with commercial success but cannot be eroded by transfer pricing

  9. Tie revenue apportionment to commercial use of models materially benefiting from Australian infrastructure, energy, public investment or copyright protections.

    That is the role of revenue apportionment, an ongoing ad valorem share tied to the commercial utilisation of models that have materially benefited from Australian infrastructure, energy, public investment or copyright protections.

  10. Calibrate public participation so investment remains commercially viable and risks and returns are allocated equitably.

    Such a multi- instrument framework must, in turn, be calibrated carefully so that public participation in the economic value generated does not render investment commercially unviable or allocate risks and returns inequitably between participants.

  11. Use an access fee as an immediate revenue floor, distinct from a tax on training or payment for works used by models.

    Beneath the noted layers in the model sits an access fee that provides an immediate revenue floor. It is worth being precise about what this fee recognises, because it is not a charge on training as an activity and still less a payment for the works a model learns from.

  12. Charge for public endowments making training possible, including land, grid capacity, electricity, water, planning and institutional stability.

    It is recognition of the public endowment that makes hosting training here possible at all, the land and greenfield sites that large facilities consume, the transmission capacity and the firmed, increasingly clean electricity drawn from a shared grid, the clean water used for cooling, the accelerated planning and approvals, and the stability of a well-governed jurisdiction.

  13. Set the access fee with reference to long-term upside; consider offsets against sufficient future revenue capture or an annual minimum charge offset against throughput revenue, potentially unnecessary with a well-designed apportionment layer.

    The fee should not be set without reference to long-term upside, because a fee fixed in isolation is a one-time charge whose inadequacy only becomes visible in retrospect. In the layered model it does not have to carry that weight, because the ownership and apportionment layers capture the upside it cannot. This fee could also be offset against future revenue capture should future revenues achieve a level of sufficiency. It is also plausible that the fee could serve as an annual minimum charge that is offset against throughput revenue and therein serving as a revenue floor, ensure no repeat of the LNG scenario, though this approach may be unnecessary with a well designed second layer.

  14. Reserve a share of sovereign compute for universities, public-interest research, start-ups, the national science agency and strategic needs, as a condition of access to Australian infrastructure.

    Reserve a share of sovereign compute in kind for universities, public-interest research, start-ups, the national science agency and strategic needs, as a condition of access to Australian infrastructure.

  15. Include uplift of the national science agency through monetary or compute allocations in the sovereign reservation.

    It is worth noting that Australia’s principal national science agency has been significantly depleted, so a reservation that includes uplift of that agency (through monetary or compute allocations) would serve a national interest beyond the AI sector itself.

  16. Quarantine proceeds in a dedicated national fund using Future Fund or Future Ireland Fund architecture, separating long-term asset value from the operating budget.

    Quarantine the proceeds in a dedicated national fund on the fiscal architecture of the Future Fund or Ireland’s Future Ireland Fund, separating long-term asset value from the operating budget.

  17. Give the national fund an investment mandate managing externally driven, potentially cyclical rents, and governance independent enough to survive changes of government and price cycles.

    Three design features are important, each of which is highly consequential. First, a clear separation between the captured revenues and the operating budget, to prevent the long- term asset value being spent as current income. Second, an investment mandate that manages the externally-driven and potentially cyclical nature of these rents, which is the lesson of Chile’s stabilisation fund if AI infrastructure rents prove volatile. Third, governance independent enough to survive changes of government and commodity-style price cycles, which is the feature that made the Norwegian arrangement durable across decades.

  18. States should consider capturing approval-related land windfalls through betterment levies or the ACT’s Lease Variation Charge, complementing the Commonwealth model rather than forming part of it.

    Land value capture is a useful front-end complement at the state level. Rezoning, grid connection and fast-tracked approval create a land windfall through the planning act itself, capturable at approval through betterment levies or the ACT’s Lease Variation Charge. It is one-time and site-bound, so it pairs naturally with the ongoing revenue and ownership mechanisms, and because it is a state lever the value accrues outside any Commonwealth fund. It is noted here for state-level consideration rather than as part of the core Commonwealth model.

  19. Design creative-sector support for ongoing structural displacement, distinct from and additional to copyright licensing, funded partly from value-capture proceeds.

    The recommendation is therefore to design a creator-support mechanism for ongoing structural displacement, distinct from and additional to copyright licensing, funded in part from the value-capture proceeds, so that the sector most exposed to AI shares directly in the dividend its work helps generate.

  20. Make displacement support more ambitious, independent and adaptable than the Public Lending Right precedent, reaching beyond named creators to journalists, designers, translators and the broader cultural economy.

    A response calibrated to the actual scale of disruption would need to be more ambitious, more independent and more adaptable, and it would need to reach beyond named creators to the broader cultural economy of journalists, designers, translators and others whose markets AI is restructuring.

  21. Encourage further exploration of attribution science, recognising its emerging evidence base; treat it as a future horizon rather than a prerequisite to acting now.

    It is acknowledged that much of the literature has centred around other forms of Machine Learning and Deep Learning and the field is an emergent one, but the shared literature provides and evidentiary basis for further exploration, something that should be encouraged. It is included as a horizon for the model rather than a precondition for acting, what is proposed within the appendices does not preclude acting now with a simpler approach.

  22. Attach national-interest conditions to training: training-run notification, model architecture and training-data transparency, and requirements governing where Australian data is stored and processed.

    The privilege of training on Australia’s advantages should carry conditions that reflect the national interest. The available conditions beyond baseline environmental and planning requirements include notification of training runs, transparency about model architecture and training-data composition, data-sovereignty requirements about where Australian data is stored and processed, prohibitions on uses contrary to Australian law or national security, a commitment to developing local talent and the skilled trades workforce, and requirements to contribute to domestic AI research capacity.

  23. Include training conditions prohibiting uses contrary to Australian law or national security and requiring local talent, skilled-trades development and contributions to domestic AI research capacity.

    The privilege of training on Australia’s advantages should carry conditions that reflect the national interest. The available conditions beyond baseline environmental and planning requirements include notification of training runs, transparency about model architecture and training-data composition, data-sovereignty requirements about where Australian data is stored and processed, prohibitions on uses contrary to Australian law or national security, a commitment to developing local talent and the skilled trades workforce, and requirements to contribute to domestic AI research capacity.

  24. Design enforceable conditions to withstand diplomatic pressure by anchoring them to fixed assets and countable units within Australian jurisdiction.

    The hardest challenge is enforceability against large global firms with limited Australian legal presence, including the risk of diplomatic pressure under Section 301 of the United States Trade Act. Conditions must be designed to withstand that pressure, which is a further argument for anchoring the regime to fixed assets and countable units within jurisdiction rather than to activity that can be relocated or restructured.

  25. Build on the skills baseline by considering an apprentice-to-tradesperson ratio tied to grid-connection and planning approvals, with settings developed alongside training bodies and unions and matched by public training investment.

    Build on the legislated skills baseline with a genuine commitment to local talent and apprentices. With training and apprenticeship requirements set to become part of the mandatory data centre standards, the question is how to make them meaningful; an apprentice-to- tradesperson ratio attached to grid-connection and planning approvals is one option worth serious consideration, with the precise settings developed in dialogue with training entities, the TAFE sector, group training organisations and unions, and matched by public investment in training capacity.

  26. Require infrastructure beneficiaries to train and employ Australian apprentices and invest in local skills, including renewable-energy projects connected to their facilities.

    The condition should be a genuine training obligation, not a box-ticking exercise, operators benefiting from access to Australian infrastructure should commit to train and employ Australian apprentices and to invest in the local skills base, including on the renewable-energy projects connected to their facilities.

  27. Consider an apprentice ratio, illustratively one to five qualified trades workers, prioritising first- and second-year apprentices to create new entrants rather than poach experienced apprentices from small businesses.

    One workable mechanism is an apprentice-to-tradesperson ratio; for example, one apprentice for every five qualified trades workers, with particular attention to first and second-year apprentices so that the obligation creates genuinely new entrants rather than drawing experienced apprentices away from small businesses.

  28. Develop precise ratios, verification and responses to shortfalls, including training contributions or alternatives, with training entities, TAFE, group training organisations and unions.

    The right settings/calibration, including the precise ratio, how it is verified, and whether shortfalls are addressed through a training contribution or other means, should be developed in close dialogue with training entities, the TAFE sector, group training organisations and unions, because those bodies hold the practical knowledge of where training capacity exists and where it must be built.

  29. Allow training commitments through an apprentice ratio or contributions to a fund supporting site-based apprenticeships and the surrounding training ecosystem.

    This is achieved by attaching a training and apprenticeship commitment to the same access that the other layers price, operators benefiting from Australian infrastructure contribute to developing the local skilled workforce, whether through a set apprentice-to- tradesperson ratio, or a contribution to a training fund that supports site-based apprenticeships and the surrounding training ecosystem.

  30. Settle access fees and reserved compute early while negotiating ownership and apportionment as investment terms, when Australian leverage is strongest.

    The layered model is designed to make ambition negotiable by sequencing it; the access fee and reserved compute can be settled early, while the ownership stakes and apportionment are negotiated as part of the investment terms, when Australia’s leverage is at its peak.

  31. Have regulators monitor attribution science for future attribution and discussions beyond copyright.

    We acknowledge that these matters are beyond the scope of the current report but must be monitored by regulators as a source of future attribution, and for discussions beyond copyright.

  32. Consider both human intellectual property incorporated during model development and downstream redistribution of economic value through AI’s share of finite attention.

    For this reason, policy responses to generative AI should consider both stages of the AI value chain. The first concerns the incorporation of human intellectual property into AI capability during model development. The second concerns the downstream redistribution of economic value arising from AI’s increasing share of finite human attention.

Read the original

In some quotes a program corrected obvious text-recognition errors for display.

Open in the interactive guide