Artificial intelligence was not built from nothing. It was built on the collective output of the world's literate populations — every article, every book, every scientific paper, every line of code, every photograph, every conversation on the open internet. It was extracted without compensation. A handful of foreign companies scraped the open internet, trained commercial products on what they found, and now sell those products back to the populations whose work made them possible. Whether that constitutes theft is a question the courts have yet to settle. Whether it constitutes a debt is not.
Australia is not unique in contributing to this. Every country with a presence on the open web has had its output ingested into the training corpus; every literate population is in there somewhere. But Australia has standing the way a coastal state has standing over its fisheries — not because the fish are Australian, but because commercial extraction from the waters is. And the extraction is not a past event. Every day an Australian opens a subscription, asks a model a question, integrates an AI tool into their work, the commons is drawn on again. Every month the industry operates commercially in this country, it draws on what Australia contributes. The question the Dividend answers is not whether Australians are owed compensation for what was taken once. It is what the terms should be for what continues to be taken, every day, from here.
The turbulence is already beginning. Journalists, graphic designers, illustrators, translators, junior coders, customer service workers — occupations that contributed their collective output to the open web for two decades, and whose output trained the very models now automating their work. The training corpus and the displacement list are the same list.
How will governments respond? There are two options on the table, and both are bad.
The first is to tax AI companies directly. Corporate income taxes. Digital services taxes. Sectoral levies. Every one of those instruments has the same structural flaw: it taxes the industry, and the industry responds by raising end-user prices. Subscriptions get more expensive. The Australians first priced out of AI are the ones who can least afford to be — workers displaced by AI-driven productivity gains, single-income households where a subscription is a real trade-off, young people entering the workforce without capital to buy the tools they need to compete. Taxation as currently conceived transfers the cost from the industry to the most exposed citizens. That is not a policy outcome. It is a policy failure, imposed on the people a policy is meant to protect.
The second is to do nothing. Let the productivity gains flow offshore. Let Australian IP continue to underwrite an industry that returns no value to the country that helped build it. This is the default option. It is also, quietly, the most expensive option Australia can choose — because the cost is the country's share of the transition, forever.
Tax AI, and the poorest pay. Don't tax AI, and Australia gets nothing back for its share of the commons. That is the dilemma every government now faces. It is the dilemma the Australian AI Dividend is designed to resolve.
Within its opening years, the Australian AI Dividend pays for a mid-tier AI subscription outright for every Australian adult — meaning universal access to the transition from the opening years onward. At maturity, it channels approximately A$4 billion per month to Australian households, roughly A$200 per adult, every month, with Australian children accumulating a share across their childhood, delivered as a lump sum at 21 — a generational inheritance from the industry their generation helped build.
The mechanism: charge AI companies for the commercial supply of AI services to Australians, measured by the energy footprint of that supply — the unit the industry is already disclosing under EU and ISO regimes. Australian startups, academic research, and open-source projects pay nothing. The largest hyperscale operators pay the highest rate. The architecture uses tools the Commonwealth already owns. No new constitutional territory. No discriminatory tax architecture.
The Australian AI Dividend doesn't work like a conventional tax, and that's the point. Conventional taxes are assessed on revenue, so providers raise prices to protect their margin and the cost lands on customers. The Dividend is assessed on energy — the kilowatt-hours burned serving Australian users. What a provider charges its customers has no bearing on what it owes. Pricing is a competitive-market decision, set against every other provider in the Australian market; the levy is set by the meter. The two are independent.
That independence is what makes universal access possible. Power users — businesses, developers, high-volume commercial integrators — will pay for the AI they consume, as they do now. But for the ordinary Australian who wants a capable AI assistant for everyday use, the monthly Dividend covers it. At current Australian consumer pricing, Grok runs A$12 a month and Google AI Plus A$12.99, with Microsoft Copilot at A$16, ChatGPT Plus at around A$30, and Claude Pro at A$34. Within the scheme's opening years, the Dividend covers Grok and Google AI Plus outright for every Australian adult. As the scheme matures, it covers any of them. Average Australians aren't charged anything extra; they get the tools for free, funded by the industry's heaviest users. The Australian AI Dividend turns AI from a subscription into a public utility.
And the AI industry itself has reason to welcome it. The alternative — conventional corporate taxation, digital services taxes, sectoral levies — delivers neither stability nor predictability. It delivers a decade of fragmented rule-making across jurisdictions, each instrument fought in turn, each outcome uncertain. The Dividend offers the opposite: one federal framework, set once and indexed over time, built on legal architecture that has worked in Australia without controversy for forty years. For OpenAI, Anthropic, Google, and Microsoft, Australia becomes the jurisdiction that got the architecture right — not the worst thing that can happen here, the best thing that can happen here.
A working policy brief, offered for consultation. Feedback from stakeholders in trade, industry, Treasury, and the legal academy is actively sought. Full legal framework, mechanism design, economic modelling, governance architecture, and responses to anticipated attacks follow.
The productivity gains from AI are real and they are large. They are also, at this moment, entirely foreign-owned. Every Australian who uses AI pays a subscription to a company headquartered abroad. Every Australian business that integrates AI into its operations transfers value to foreign shareholders. Every Australian child learning with AI is generating economic return for people who will never set foot in this country. This is not a hypothetical future. It is what is happening now, and what will happen on a larger scale every year from here.
The trajectory is clear. By 2035, a small number of foreign-owned AI systems will have reshaped Australian work, commerce, and daily life. The Australian economy will run on AI that Australians do not own. The productivity gains will be real. The returns will be offshore. And unlike every previous technological transition in Australian history — from the mining boom to the digital economy to the rise of platform companies — there is no mechanism currently in place to ensure Australians share in the value their country helps create.
The question is not whether this transition happens. It is happening. The question is who owns the value it creates. Three answers are on the table.
Do nothing. Every dollar of Australian-generated AI surplus flows to foreign shareholders. No Australian household has any direct participation in the growth of an industry that will increasingly define Australian economic life. This is the default. This is where current trajectory leads. It is also, quietly, the most expensive option Australia can choose — because it costs the country its share of the transition, forever.
Regulate sectorally, on the European model. A decade of contested rulemaking. Licensing regimes for this industry, restrictions for that one. New rules each time a model improves. Old rules fought again at every revision. Slow. Painful. Prone to capture. And — crucially — with no positive economic return for citizens. Regulation distributes cost, not benefit.
The Dividend delivers on the ownership commitment through a market mechanism. It does not regulate AI. It does not slow AI. It does not pick winners. It charges AI companies for the commercial supply of their services to Australians — assessed against the energy footprint of that supply — and returns the proceeds to the citizens on whose commons the industry draws. It arrives with the emerging international measurement regime, not against it. It uses tools the Commonwealth already owns. It rests on legal architecture that has worked, without controversy, for forty years.
This is a generational decision. Australia can be the jurisdiction that builds the template democracies use to share in the AI transition — or it can be a customer of whatever template emerges elsewhere. The window to choose is this decade. The cost of choosing late is the share itself.
This is not a levy to punish. It is a levy to participate.
Every charge needs a base — a measurable, defensible unit against which the amount owed is calculated. The choice of base shapes everything that follows: what can be gamed, what can be hidden, what can be attacked under trade law, what aligns incentives correctly. Before walking through the mechanism, we owe the reader an answer to the question any serious reader will ask first: what is being measured, and why is it the right thing to measure?
The object of the charge is the commercial supply of AI services to Australian users. The unit of assessment — the base against which liability is calculated — is the energy footprint of that supply. This is the unit the industry is already converging on globally for its own sustainability reporting. Australia arrives as an adopter of the emerging international regime, not an inventor of a new one.
The conceptual step is this. Australia's commons contribution — data, language, users, cultural output, scientific infrastructure, public research investment — is distributed across the full operation of every AI provider commercially serving Australians. No clean unit-by-unit allocation of commons draw to any individual provider is possible, and none is necessary. What is measurable is the scale at which each provider operates commercially — and at commercial scale, operation necessarily draws on the commons. Energy footprint is the disclosed, verifiable, internationally-aligned unit by which that scale is captured. The charge is proportional not to each provider's specific commons extraction (impractical), but to the scale of commercial activity that necessarily draws on the commons (practical and legally defensible).
Industry already discloses the unit. Under the EU AI Act, providers of general-purpose AI models have been required to document the energy consumption of their models since 2 August 2025 — training energy at the model level, disclosed via the Annex XI technical documentation regime. Google published its Gemini per-prompt energy methodology in August 2025, reporting 0.24 watt-hours per median text query and establishing a de facto reference for inference-side disclosure. ISO/IEC Technical Report 20226, covering environmental sustainability aspects of AI systems, was published in July 2025, with full International Standards on AI training efficiency and sustainability reporting scheduled through 2026–2028. The unit — energy footprint — is the measure the international regime is converging on for AI operational impact.
What the international regime does not yet do is apportion that energy to specific jurisdictions. Training energy is disclosed per model, globally. Per-query figures are disclosed as global averages. The Australian share of each provider's inference energy is not a figure any international body currently publishes. That apportionment is the work Australia does — the same kind of architectural move Australia made in 2017 when it extended GST to imported digital services. The global principle was established internationally; the Australian application was built here. The Dividend follows the same pattern: adopt the international unit, extend it with Australian apportionment methodology, phase in direct measurement as the international regime matures.
For the scheme's opening years, the Dividend Authority publishes deemed rates — administrative estimates of Australian inference energy derived from providers' global disclosures and Australian market factors. This is not a placeholder while waiting for the international regime to catch up. It is the scheme's initial assessment method, with direct Australian precedent in FBT statutory formulas and fuel tax credits, both of which have operated on deemed-rate principles for decades. As the international disclosure regime matures and Australian apportionment methodology is piloted and refined, the scheme transitions to direct measurement — per §03 Mechanism's three-phase assessment regime.
It scales with commercial activity. More supply to Australian users means more energy consumed by that supply. The unit grows with the thing being measured. A provider who serves twice as much Australian inference as another pays twice as much. There is no gaming opportunity in the unit itself.
It is externally verifiable. Unlike tokens (proprietary, provider-specific), user counts (gameable, re-definable), or revenue (offshorable, transfer-priceable), energy disclosure is becoming internationally auditable under EU and ISO regimes. Australia relies on the same disclosed numbers that apply in other jurisdictions — the verification infrastructure is being built by the world, not by Canberra.
It is provider-neutral. The same unit applied identically to every provider supplying Australians. Australian and foreign providers face identical assessment on identical inputs. No discriminatory treatment. No Digital Services Tax problem. No preferential domestic carve-outs.
Most AI serving Australians today is inferenced in US hyperscaler data centres, and that will still be substantially true in 2030. This does not weaken the scheme — it is why the scheme's foundation is not Australian electricity but Australia's contribution to the global AI commons. Energy is the unit by which AI companies' commons-derived revenue is measured. The obligation is to Australia, whose data, users, language, cultural output, and public research investment make the commercial supply possible. The physical location of electrons is incidental.
Put plainly: the electrons don't have to be Australian. The obligation does.
This distinction resolves the attack that breaks every previous version of this kind of proposal. The scheme is not claiming jurisdiction over US electricity. It is charging AI companies for supplying Australian users commercially, and using energy footprint as the internationally-accepted unit of measure for that supply. Foreign providers register, self-disclose under the international regime, and submit to audit — exactly as Netflix, Spotify, and other foreign digital suppliers have done under the Australian GST on imported digital services since 2017.
Three components. Each is simple; together they close the loop.
AI companies supplying Australian users pay a charge on the commercial supply of AI services to those users. The amount owed is assessed against the energy footprint of that supply, using the international disclosure regime. The charge is progressive and banded. The smallest suppliers — Australian startups, academic research, open-source projects — pay nothing. The largest hyperscale operators pay the highest marginal rate.
| Annual Australian inference energy | Marginal rate |
|---|---|
| First 1 GWh | Nil |
| 1 – 10 GWh | 5¢ / kWh |
| 10 – 100 GWh | 15¢ / kWh |
| 100 – 500 GWh | 30¢ / kWh |
| 500 – 1,000 GWh | 50¢ / kWh |
| Above 1,000 GWh | 75¢ / kWh |
Measurement follows the emerging international regime — the EU AI Act (energy disclosure obligations for GPAI providers operational since August 2025), Google's published Gemini per-query methodology, and the ISO/IEC standard for AI sustainability. The assessment regime moves in three phases: regulator-published deemed rates (years 1–3), hybrid measurement (years 3–7), direct metering based on provider disclosure thereafter.
Foreign providers supplying Australian users register with the Dividend Authority as Registered Suppliers, disclose energy footprint figures under the international regime, and submit to audit by the Australian Taxation Office. This parallels the existing GST on imported digital services regime — the "Netflix Tax" — which has handled foreign digital suppliers' Australian obligations operationally since July 2017. Enforcement draws on Australian tax administration's established audit capacity, with direct precedent in the Petroleum Resource Rent Tax, the fringe-benefits-tax statutory formula methods, and fuel tax credits.
Revenue from the levy is distributed monthly to every Australian adult as Dividend credit. Credit is denominated in Australian dollars at face value — one credit is one dollar — and held in a citizen account administered by the Dividend Authority. Eligibility is Australian residency for tax purposes, age 18 and over. Credit lands on the first of each month; whatever is not spent or transferred by month's end sweeps automatically to the Endowment.
Citizens can do three things with their credit. They can spend it on AI compute at any Registered AI Provider. They can transfer it to another eligible Australian adult, or to a dependent child account held under their own control. They can let it sweep to the Endowment, with the option to contribute voluntarily at any point in the month rather than waiting for the automatic sweep.
Redemption at AI providers is direct. A citizen uses their AI subscription or service as they would today; the provider bills the Authority's holding fund in Australian dollars for the compute that citizen consumed; the citizen's balance is drawn down by the amount billed. If a citizen has fifteen credits and their month's AI usage costs $12.90, their balance after billing is $2.10, which sweeps to the Endowment at month's end. There is no parallel scrip, no exchange rate, no spot price. One credit is one dollar of AI compute at the provider's ordinary Australian price.
Peer transfer — P2P — is the scheme's social mechanism. An eligible adult may transfer any portion of their balance to any other eligible adult, or to a dependent child account they control. There is no cap on transfers, and transferred credit can be transferred onward by the recipient. A grandparent topping up a grandchild's study tools, a parent funding a teenager's AI-assisted schoolwork, a friend gifting credit to a power user who will put it to use — all are first-class operations within the ecosystem. The Dividend legislation provides that transfers of Dividend credit between natural persons are not taxable supplies; P2P transfer carries no GST consequence.
Only natural persons hold credit. No companies, trusts, charities, or estates. If a credit-holder dies, any balance sweeps to the Endowment. Child accounts exist solely under parent control — the parent distributes, revokes, and governs the account's activity. Child accounts may only redeem credit at AI providers whose terms of service admit the age of the account-holder; the Authority does not override provider age policies.
Registered AI Providers are the levy payers themselves. Any provider paying into the Dividend Fund above the first gigawatt-hour is automatically eligible to receive credit redemptions — one register, one rule. Sub-de-minimis providers (Australian startups, academic research, open-source projects) neither pay the levy nor participate in redemption; they operate outside the scheme on both sides, as intended. The Authority maintains a citizen-facing directory of Registered Providers; a citizen sees which services their credit can be used at and engages with those services as they would today.
The scheme assumes digital engagement. Redeeming credit at an AI provider is an online action by nature, and citizens who cannot engage digitally with an AI service would not be able to consume what the credit buys regardless of the access mechanism. For those citizens, the two available paths are voluntary contribution to the Endowment — a direct intergenerational gift — and P2P transfer from a family member who holds the account on their behalf. Both are first-class uses of the credit.
The architectural consequence: no arbitrage vector, no parallel market, no secondary valuation. Credit redeems at face value at a defined set of providers, transfers among eligible persons, or accumulates in the Endowment for Australian twenty-one-year-olds. Every flow is visible, every transaction is in Australian dollars, every participant is a natural person or a registered AI provider paying into the fund.
Unredeemed credit flows to the Dividend Endowment — a Commonwealth statutory body modelled directly on the Future Fund Act 2006. The Endowment invests, accumulates, and pays out a single lump sum to each Australian at age 21, calculated by residency-weighted share (days of eligible residency over total cohort residency days).
A Future Generations Advisory Panel — Australians aged 18–25 — sits in statute as a formal consultative voice on decisions affecting future beneficiaries. Core provisions are entrenched through absolute-majority amendment requirement and a mandatory 180-day public consultation period. Amendable under serious deliberation. Protected from casual rollback.
Value enters the system through providers paying the levy. It distributes to Australian adults as monthly credit. From there it moves in three directions — and this is the design's central feature.
Credit spent at a Registered AI Provider buys compute at face value. A citizen uses the AI service as they would today; the provider bills the Authority's holding fund in Australian dollars for what was consumed; the citizen's balance draws down accordingly. For the ordinary Australian wanting a capable AI assistant, the monthly Dividend covers a mid-tier subscription outright within the scheme's opening years and any of the premium tiers as the scheme matures. No exchange rate, no scrip, no parallel unit of account — one credit, one dollar, one ordinary commercial transaction with a sovereign payer.
Credit transferred between adults is the scheme's social mechanism. A grandparent tops up a grandchild's study tools. A parent funds a teenager's AI-assisted schoolwork through a child account. A friend gifts credit to a power user who will put it to work. The transfer carries no tax consequence and no cap; it is person-to-person, intra-family or inter-generational, as Australians already handle small sums of money among one another. The difference is that the sum originates with the Commonwealth, funded by the industry drawing on the commons — and moves among Australians before it is either used or returned to the next generation.
Credit not spent or transferred by month's end sweeps to the Dividend Endowment. This is the scheme's intergenerational engine. Every Australian adult who does not use AI, does not transfer, and does not choose to contribute voluntarily still participates — their share compounds in a sovereign fund that pays out to every Australian at twenty-one. The effect compounds: the less a cohort uses AI today, the greater the inheritance waiting for the next cohort starting out.
The loops reinforce. Provider revenue drives inference volume, which drives levy receipts, which drives credit issuance. Citizen redemption rewards providers who serve Australian users well. P2P transfer keeps credit moving among Australians who value the scheme's use by someone they know. Unspent credit compounds in the Endowment and releases as capital to young Australians at the age where a lump sum most changes a life trajectory. Every flow is visible. Every participant is a natural person or a registered provider paying into the fund. No cash exit. No parallel market. No arbitrage vector.
Every stakeholder pays something and gets something greater. That is the engineering test the design was built to pass.
Figures are working assumptions. Formal calibration is Treasury and Parliamentary Budget Office work.
What moves the trajectory within the ranges. Three working assumptions determine where within each range the scheme actually lands: industry inference growth, band-calibration accuracy, and citizen redemption rate. Faster industry growth and higher redemption shift the per-adult figure upward within the range; slower growth and lower redemption shift it down. Formal Treasury and Parliamentary Budget Office calibration against real industry data will resolve these ranges to narrower point estimates. Until that work is complete, ranges are the honest way to describe what the scheme delivers.
It is tempting to read the early-year figures — $10–$20, $35–$65 — as modest. They are, in absolute terms. The framing misses the point.
At present Australian consumer pricing, a mid-tier AI subscription — Grok at A$12, Google AI Plus at A$12.99, ChatGPT Go at A$12.99, Microsoft Copilot at A$16 — is fully covered once the Dividend crosses that threshold, which working assumptions place within the scheme's opening years. Premium subscriptions — ChatGPT Plus around A$30, Claude Pro A$34 — are substantially subsidised at that point and fully covered as the scheme matures. For the median Australian — someone who wants AI to help write an email, plan a trip, brainstorm a gift for their partner's birthday, sort a recipe — the mid-tier tools are capable and the Dividend covers them in full once it reaches that level. No out-of-pocket cost.
The industry does not make its money from that user. The industry makes its money from power users: businesses, developers, creative professionals, high-volume commercial applications. Those users pay many hundreds of dollars a month; their personal Dividend is a rounding error in their AI spend. The scheme is engineered around this asymmetry. The light user is substantially funded by the Dividend. The heavy commercial user contributes in proportion to their consumption, as they do today.
The structural effect is universal participation without universal subsidy. Every Australian adult participates in the scheme: through AI access directly, through P2P transfer to someone who will use it, or through Endowment contribution for the next generation. No Australian is left out, and no Australian is forced in. The industry's revenue profile is unchanged. The funding comes from the same commercial users already paying at scale.
By the time the Dividend reaches $100–$200 per month in later years, its purpose evolves — from "universal entry to AI" to "meaningful monthly household supplement." But the first-years case is not that the Dividend is large. The case is that it is exactly the right size to buy every Australian entry to a transition that would otherwise pass them by.
Unredeemed dividend compounds over decades. An Australian child born in 2030 accumulates twenty-one years of endowment contribution before payout. The target is a lump sum meaningful enough to change life trajectories for Australians of modest means — comparable in magnitude to a first-home deposit or a substantial education fund — but not so large as to be distortionary.
Precise figures depend on redemption rate, investment return, population growth, and the industry's energy trajectory. Formal modelling belongs to Treasury and PBO. The design constraint is clear: the payout should make a real difference in the life of a twenty-one-year-old starting out — and no more.
The top marginal rate — 75¢ per kWh — applies only to inference energy above 1,000 GWh per year served to Australian users. That is true hyperscale operation. The band exists for the maturing market: it indexes Australia's future capture to the industry's actual growth rather than its current size. The rate structure rewards the industry for scaling.
The proposal is positive-sum by construction. Each stakeholder gains something specific, and the gains reinforce.
A meaningful monthly dividend — ten to twenty dollars in the early years, one to two hundred dollars per month at maturity. Usable at AI providers for compute, transferable to another Australian adult or a dependent child account, or held for later. Real economic participation in productivity gains currently captured entirely offshore. A share of the AI future that does not depend on employment by a foreign-owned vendor. For the median Australian, the early-years Dividend buys a mid-tier AI subscription outright and substantially subsidises premium tools — a seat at the table, paid for by the industry itself.
A sovereign endowment accumulating across their childhood, delivering a lump sum at age twenty-one. Not a universal basic income. Not a means-tested welfare payment. An intergenerational compact that treats AI productivity as national inheritance — delivered at the age where a capital base most changes a life: starting a business, completing study, entering the housing market. A generational asset, built while they grew up, released as they start out.
Regulatory stability. One federal market mechanism, set once and indexed over time, is substantially preferable to the sectoral EU model or fragmented state-level action. Market expansion: every Australian adult has subsidised access to participating services, which grows the addressable market rather than fragmenting it. De minimis protection for startups, academia, and open-source research — the first gigawatt-hour is free. Australia becomes a predictable jurisdiction, not a hostile one. This is not a punitive regime. It is a durable one.
A political buffer against AI-displacement politics — a visible, durable citizen stake that changes the political equation when sectoral regulation is proposed. A sovereign asset independent of commodity cycles. Demonstrable sovereignty without protectionism, aligned with rather than divergent from emerging international frameworks. A durable fiscal mechanism that grows with the industry it levies.
The proposal has been stress-tested against ten distinct attack vectors from across the political spectrum. The six most likely to arise in early engagement are answered here. The remaining four — fig-leaf framing from the left, accessibility concerns, Greens bundling with sectoral regulation, and author credentials — have established rebuttals in the mechanism design and the full proposal.
The usual first objection, answered in two parts. The unit is internationally settled: the EU AI Act has required general-purpose AI providers to disclose training energy since August 2025, Google published its Gemini per-prompt inference methodology in August 2025 (0.24 watt-hours per median text query), and ISO/IEC TR 20226 on AI environmental sustainability was published in July 2025. By the Dividend's 2030 launch, the international regime will have been operational for five years and the first full International Standards on AI training efficiency and sustainability reporting will have followed. The question is not whether energy can be measured — it is being measured, by the industry itself, under internationally agreed frameworks.
The apportionment work — allocating each provider's global inference energy to the Australian share — is built here, by Australia, using the same kind of architectural move that extended GST to imported digital services in 2017. For the scheme's opening years, the Dividend Authority publishes deemed rates derived from providers' global disclosures and Australian market factors, with direct precedent in the FBT statutory formulas and fuel tax credits that have operated on deemed-rate principles for decades. The scheme transitions to direct measurement as the international regime matures. Enforcement uses the Australian Taxation Office's established audit capacity, with direct precedent in PRRT. For the minority onshore case, AEMO and NGER already capture facility-level energy data. The architecture is formulaic, internationally aligned, and externally auditable.
The Dividend is not a Digital Services Tax. DSTs tax the gross receipts of foreign firms from services supplied to domestic users — the structure the OECD has spent a decade disputing, and the structure that saw Canada's DST withdrawn under US pressure in 2025. The Dividend charges all suppliers, domestic and foreign, identical rates assessed against an industry-disclosed, internationally-aligned unit of measure. There is no preferential treatment of Australian suppliers and no offshore loophole.
The deeper trade-law answer is structural. A facially neutral measure can be challenged as de facto discriminatory if its burden falls disproportionately on foreign suppliers in a way that tracks nationality rather than a legitimate policy basis. The Dividend's progressive band structure is calibrated to operational scale, not to nationality. At launch, the top band will in practice be populated predominantly by foreign hyperscalers — because hyperscale AI infrastructure is currently concentrated in a small number of globally-capitalised firms, and that concentration reflects international capital-market reality, not Australian policy. The band structure charges the scale, whoever produces it. Over the scheme's lifetime, as AI infrastructure scales and the threshold reaches further down the industry, the top band captures any serious commercial participant in the Australian market — foreign, domestic, or mixed. The structure's policy basis is commons-extraction intensity measured by operational scale, which is a legitimate and long-recognised basis for progressive charging.
The operational model for foreign suppliers is the GST on imported digital services — the "Netflix Tax" — which has handled Australian obligations on Netflix, Spotify, Apple, and hundreds of foreign digital firms since July 2017, without trade dispute. The Dividend extends this existing, operational, uncontroversial Australian regime to a new domain, grounded in the Commons Doctrine: Australia's contribution to AI, through its data, users, and infrastructure, establishes sovereign authority to charge for the commercial supply drawing on that commons. Trade law experts will recognise the structure on sight.
Universal AI access is the Australian value the Dividend delivers. The productivity transition is the largest economic shift of the century; a capability gap between Australians and populations with unrestricted AI access is a permanent economic disadvantage. The alternative — an Australian population priced out of the tools everyone else is using to compete — is precisely the outcome the scheme is designed to prevent. The Dividend does not subsidise foreign AI; it secures Australian access to a transition underway regardless of whether Australia participates.
The Endowment sweep is the scheme's answer to the circulation question. Credit that Australians do not spend on AI or transfer among themselves compounds in a sovereign fund, decade after decade, and pays out to Australian twenty-one-year-olds. The slower the redemption rate, the larger the Endowment's growth — every month a cohort underuses AI, the next cohort's inheritance grows. The scheme's economic geography is not a round trip. It is a bifurcation: immediate productivity access for users, sovereign capital accumulation for the generation starting out.
The Dividend also does not foreclose other policy instruments. Infrastructure investment, sovereign AI capability, research funding, industrial policy — all remain available, funded through the ordinary Commonwealth budget process. The Dividend addresses one specific question (how Australia captures value from commercial AI extraction) and leaves every other question of AI policy available to be answered on its own terms.
The Dividend rests on a specific legal theory: Australia contributes materially to the operation of modern AI — through its data, its users, its infrastructure, its languages, its cultural and scholarly output, and sixty years of public investment in science, education, and communications — and the Commonwealth exercises sovereign authority over that contribution. The levy is a charge for the commercial supply of AI services drawing on that contribution, assessed against the energy footprint of that supply. It is not a market intervention. It is not a Digital Services Tax, and it is not a tax on foreign activity in the discriminatory sense that would engage trade-law obligations.
Australia's data, user base, language, cultural output, scientific infrastructure, and six decades of public research investment together constitute a commons from which every commercial AI operation materially draws. Training corpora scraped from the open web include Australian contributions. Model performance on Australian English depends on Australian language data. Commercial viability in the Australian market depends on Australian users choosing to adopt the product. These are not incidental inputs — they are conditions on which commercial AI operation in Australia is possible.
The Commonwealth exercises sovereign authority over this commons on the same constitutional footing it has exercised authority over other national commons for more than a century — minerals, petroleum reserves, radio spectrum, territorial waters, fisheries. The characteristic Australian policy response to commercial extraction from a commons is not prohibition, but charging at a rate that captures super-normal returns without impeding operation. The Dividend applies this established pattern to a new domain.
This theory — Australia's Share in the Global AI Commons — is held as a reserved technical doctrine, to be developed formally with senior Australian legal academic and former Commonwealth law-officer input during consultation. This brief uses the ownership framing in civic language; the doctrinal work sits behind it.
Before the constitutional analysis, a clarification worth making up front. The word tax is used in this brief — and in policy discussion generally — in two distinct senses, and they should not be confused.
In the narrow, international sense, a "tax" is a specific political and fiscal category: an income tax, a goods-and-services tax, a Digital Services Tax. These are named instruments with characteristic designs and characteristic trade-law implications. When this brief says the Dividend is not a Digital Services Tax, it is speaking in this narrow sense.
In the broad, Australian constitutional sense, a "tax" under s51(ii) is any compulsory exaction of money by a public authority for public purposes, other than a fee for service or a penalty (Matthews v Chicory Marketing Board; Air Caledonie International v Commonwealth). Under this test, PRRT is a tax. Excise duties are taxes. Customs duties are taxes. The Medicare levy is a tax. So is the Dividend. In the constitutional sense, the Dividend is a tax — and it has to be, because s51(ii) is the head of power under which instruments of this kind are authorised.
These two senses are not in conflict. The PRRT is a tax in the constitutional sense and has been for thirty-eight years; it is not a "petroleum services tax" in the DST sense, has never been trade-challenged on that basis, and no government has ever described it as one. The Dividend occupies the same architectural position: a compulsory commons-charge, authorised under s51(ii), which is not a Digital Services Tax in the specific discriminatory sense that would engage WTO, FTA, or OECD DST obligations.
The brief uses "tax" in the narrow sense when distinguishing the Dividend from a DST, and in the broad sense when discussing constitutional authority. Both are correct in their own register.
Three constitutional powers support the scheme, each long-established and uncontroversial.
Section 51(ii) of the Australian Constitution empowers the Commonwealth to levy taxation. The Dividend — a compulsory charge by a public authority for a public purpose — falls squarely within this power. The structure is identical to the Petroleum Resource Rent Tax, which has operated under s51(ii) since 1988 without constitutional challenge and without trade-law dispute. The PRRT, like the Dividend, charges commercial operators at a formulaic rate for activity drawing on a domestic contribution (Australia's offshore petroleum reserves, in PRRT's case; Australia's commons contribution to AI, in the Dividend's). Both are constitutional taxes. Neither is a Digital Services Tax.
Section 51(xxxix) — the incidental power — authorises the statutory machinery (Authority, Endowment, Advisory Panel) needed to administer and distribute the proceeds. Section 81, the appropriations power, governs the payment of dividend credits to citizens, on the same constitutional footing as every other Commonwealth disbursement.
Each of these powers has been tested repeatedly in the High Court. The scheme invents no new constitutional territory.
Four established Australian precedents provide direct architectural support. Each is embedded in Commonwealth statute and has operated for years or decades without controversy. A fifth element — the tax treatment of peer-to-peer Dividend credit transfers — is handled through a bespoke statutory carve-out in the Dividend legislation itself, on the same pattern Parliament used when it established the GST regime for imported digital services in 2017.
The PRRT — introduced by the Hawke-Keating Labor government, operating continuously since 15 January 1988 — establishes that the Commonwealth may charge for commercial exploitation that draws on a domestic contribution, using a formula that captures super-normal returns without interfering in commercial operation. Administered by the Australian Taxation Office. Thirty-eight years of continuous operation. No trade-treaty dispute. No constitutional challenge. Different commodity, same constitutional principle.
Since 1 July 2017, foreign suppliers of digital services to Australian consumers — Netflix, Spotify, Apple, Google, and hundreds of smaller firms — have been required to register with the Australian Taxation Office, self-disclose their Australian sales, and remit GST. The regime is operational, uncontroversial, and has generated billions of dollars in revenue without trade dispute. The Dividend extends this existing pattern to a new domain: foreign AI suppliers register as Registered Suppliers with the Dividend Authority, disclose energy footprint under the international regime, and submit to ATO audit. The legal and administrative infrastructure is established. The Dividend adds subject matter, not architecture.
Royal Assent 23 March 2006. The Act establishes the architecture for a Commonwealth-owned investment vehicle: an independent Board of Guardians (chair plus six members, five-year terms, statutorily required qualifications in investment, law, or corporate governance); a supporting management agency; a statutory Investment Mandate issued by the responsible Ministers; independence from government in investment decisions; and Commonwealth ownership of the fund at all times.
The Dividend Endowment is constituted directly on this template. In 2026, the Future Fund Board of Guardians model is the most proven piece of Commonwealth investment governance in existence — a sovereign wealth manager operating at institutional-investor scale, signed to the Santiago Principles on sovereign wealth fund governance, and responsible today for the Future Fund, the Medical Research Future Fund, the DisabilityCare Australia Fund, and the Housing Australia Future Fund.
The ACMA spectrum licensing regime and ASIC foreign provider registration framework establish that the Commonwealth may condition market access on registration, compliance, and reporting — without raising constitutional or trade-law concerns. The Dividend's Registered Supplier requirement (AI providers must register to supply Australian users commercially) sits within this established pattern.
The Dividend is designed to satisfy Australia's trade-treaty obligations. The analysis is straightforward on the facial-discrimination question and substantive on the de facto question — both answers are given here; formal trade-law review by senior practitioners is reserved work for consultation.
Under the WTO, the General Agreement on Trade in Services, ANZTEC, DEPA, the Australia–US FTA, and the JAEPA, Australia is obliged to avoid discriminatory treatment of foreign service suppliers. The Dividend applies identical rates to every supplier — Australian and foreign — assessed against an internationally-disclosed unit of measure. There is no preferential treatment for Australian suppliers. There is no targeting of foreign commercial activity. There is no offshore loophole. On the facial-neutrality question, the answer is clean.
The substantive question is de facto discrimination: whether a facially neutral measure can be challenged because its burden falls disproportionately on foreign suppliers. WTO and FTA jurisprudence asks whether that disproportionate burden tracks nationality or tracks a neutral characteristic with a legitimate policy basis. The Dividend's progressive band structure tracks operational scale. At the scheme's launch, the top band will in practice be populated predominantly by foreign hyperscalers, because hyperscale AI infrastructure is presently concentrated in a small number of globally-capitalised firms — a fact of international capital-market structure, not of Australian policy. The band design does not respond to nationality; it responds to scale. Over the scheme's lifetime, as AI infrastructure scales globally and the threshold reaches further into the industry, the top band captures any serious commercial participant in the Australian market, foreign or domestic. The policy basis is commons-extraction intensity measured by operational scale — the same basis on which PRRT has charged petroleum operators progressively for thirty-eight years.
Canada's Digital Services Tax, withdrawn in 2025 under US pressure, illustrates what this analysis is designed to avoid. The Canadian DST's thresholds were calibrated around revenue figures that captured US tech firms specifically, with no independent policy basis beyond the fact that foreign firms clustered above those thresholds. The structural justification was thin, and the measure did not survive. The Dividend's structure is substantively different: progressive charging on operational scale, with commons-extraction intensity as the independent policy basis, and the band structure calibrated for the industry's future shape rather than its current national composition.
The structural distinction from DSTs can also be stated in one sentence: DSTs tax the gross receipts of foreign firms from services supplied to domestic users; the Dividend charges all suppliers, Australian and foreign, for the commercial supply they make to Australians, assessed by energy footprint under the international regime. This distinction has been examined by the OECD in its multi-year work on DSTs, and the Dividend's structure falls outside the DST category on every criterion the OECD uses to identify one.
The operational model for foreign suppliers is the Netflix Tax. Foreign suppliers register, self-disclose, submit to audit, remit what's owed. Australia has been running this architecture uncontroversially since 2017. The Dividend extends the architecture to a new unit of assessment and a new domain. The trade-law objection, on examination, is answered by established facial neutrality, a coherent substantive defence of the band structure's policy basis, and an operational model already running without dispute for nine years.
No jurisdiction has yet charged AI commercial supply on commons-doctrine grounds. The EU is building out energy disclosure obligations under the AI Act but has not attached a charge to them. The UK's Digital Services Tax targets revenue rather than commons contribution and faces continuing international friction. The OECD's Pillar One and Pillar Two work addresses profit allocation, not commons extraction. The architectural move the Dividend proposes — adopt the international measurement regime, apply the charge on commons-doctrine grounds, distribute as citizen dividend and sovereign endowment — has not yet been made anywhere.
Australia is well-placed to make it. The precedents at its disposal — PRRT, the Netflix Tax, and the Future Fund Act — are among the cleanest in the Commonwealth for the architectural work this scheme requires. The window for first-mover positioning is not indefinite. The question of whether democratic states can capture value from AI at scale, on grounds defensible in trade law and durable across electoral cycles, will be settled in one direction or another within this decade. Australia can be the jurisdiction that answers it constructively, or can wait and implement whatever model emerges elsewhere.
Three bodies carry the scheme. Each has a specific constitutional function, a defined independence mechanism, and a direct precedent in existing Commonwealth governance.
The Authority is the regulator and administrator. Its functions: register providers, publish deemed energy rates, operate the audit regime, calculate and distribute monthly dividends, and report annually to Parliament.
It is structured as a portfolio agency — not a ministerial department — chosen deliberately so that regulatory decisions, especially the deemed-rate schedule, are made at arm's length from the government of the day. Independence here is not a nicety; it is the mechanism by which the scheme remains politically credible across electoral cycles.
The Authority's Commissioner is appointed by the Governor-General on the advice of the Treasurer, for a single seven-year term with statutory protection against removal except for cause. This mirrors the appointment architecture of the Commissioner of Taxation and the ACCC Chair — established Commonwealth practice for regulators requiring both independence and accountability. The Authority reports annually to Parliament, submits to Senate Estimates, and operates under the Public Governance, Performance and Accountability Act 2013.
The Endowment is a Commonwealth statutory body modelled directly on the Future Fund Act 2006. It holds and invests unredeemed credit value. It pays out to Australians at age 21 per the residency-weighted formula.
Its Board of Guardians — chair plus six members, appointed by the Treasurer and Minister for Finance for five-year terms — is constituted under the same template as the Future Fund Board. Section 38-equivalent qualifications apply: members must have substantial experience and standing in investing, investment management, corporate governance, business, or law. Members serve part-time, with remuneration set by the Remuneration Tribunal.
The Endowment operates under an Investment Mandate issued by the responsible Ministers, publicly available, amendable only through statutory process. Investment decisions are made independently by the Board; the Commonwealth retains ownership of the corpus at all times. This is the same structural pattern that has governed Australia's sovereign wealth management, without controversy, for two decades — a pattern that in 2026 oversees the Future Fund, the Medical Research Future Fund, the Disability Care Australia Fund, and the Housing Australia Future Fund.
The Advisory Panel is a statutory body of fifteen Australians aged 18–25, appointed for three-year staggered terms through a public nominations process. Its role is consultative: formal input on governance decisions affecting the Endowment's future beneficiaries, with the right to publish a dissenting report where disagreement arises.
The Panel is a voice, not a veto. Its presence in legislation is the mechanism by which the scheme maintains intergenerational legitimacy — the generation that will be the primary recipients of the Endowment has a formal seat at the table where decisions about that Endowment are made.
The choice to constitute the Endowment as a Commonwealth statutory body rather than a trust is deliberate and carries three specific legal advantages.
Constitutional cleanliness. Statutory bodies are created by Parliament under established heads of power, with clear lines of accountability to Parliament. Trust structures raise questions of trust law, trustee duties, and the identity of beneficiaries that invite litigation. The Future Fund's statutory-body structure has never been constitutionally challenged; equivalent trust structures have repeatedly faced beneficiary disputes.
Amendability. The scheme can evolve through ordinary parliamentary process (subject to the entrenchment mechanisms below). A trust would require court-supervised variation — expensive, slow, and subject to judicial discretion.
Governance transparency. Statutory bodies publish annual reports, submit to parliamentary committee scrutiny, and operate under the Public Governance, Performance and Accountability Act 2013. Trusts have no equivalent accountability architecture. For a fund that will grow to tens of billions of dollars over two decades, the accountability architecture matters.
Core provisions — the distribution formula, the age-21 payout trigger, the Advisory Panel's statutory voice, the levy's progressive structure — are protected through two parliamentary mechanisms: absolute-majority amendment requirement in both houses, and a mandatory 180-day public consultation period before any amendment may proceed.
These are ordinary parliamentary mechanisms, not constitutional entrenchment. The scheme remains amendable under serious deliberation — but not by casual legislative moment. This is the correct tier of protection for a policy that should evolve as circumstances change, but should not be quietly dismantled.
Not sooner. Four reasons, each load-bearing.
Industry growth between now and 2030 substantially increases the Dividend's per-citizen yield at launch. Launched in 2027, the scheme delivers perhaps ten dollars per adult per month initially — below the threshold at which citizens perceive it as real rather than nominal. Launched in 2030, the first month's dividend covers a mid-tier AI subscription outright for every Australian adult. That difference matters enormously: first impressions compound politically. A dividend that arrives as real is defended as real. A dividend that arrives as token is mocked and eventually repealed.
The EU AI Act's energy disclosure obligations for general-purpose AI providers entered into application on 2 August 2025 and will have been operating for five years by launch. Google's published Gemini methodology is becoming the de facto international reference. The ISO/IEC sustainability standard is expected by 2028, with Commission-issued measurement methodologies following under EU delegated acts. Launching in 2030 lets Australia adopt a mature, internationally-verified measurement regime rather than invent one. That is the right posture for a mid-sized economy. Leading on AI measurement methodology is the work of the US, the EU, and ISO — Australia's leverage is in intelligent alignment.
The coming Commonwealth electoral cycle provides a natural window for consultation and drafting. Legislation passed in 2028 gives the Authority and Endowment eighteen months to stand up operations before first distributions in 2030. This is the pace at which serious Commonwealth reforms have historically been implemented — fast enough to maintain political momentum, slow enough to get the detail right.
Operational detail — citizen-app specification and P2P transfer rails, audit procedure, endowment investment mandate calibration, deemed-rate schedule, Advisory Panel nominations process — must be resolved carefully. The lead time allows each of these to be worked out with stakeholders, piloted against real operators, and corrected, rather than improvised at launch.
2026–2027. Public consultation. Formal development of the Commons Doctrine with senior legal academics and former Commonwealth law officers. Parliamentary Budget Office costing. Targeted engagement with AI providers, Treasury, and the Australian Taxation Office. Industry roundtables under Chatham House rules to test design assumptions.
2028. Legislation drafted, introduced, and passed. The Dividend Authority Act and the Dividend Endowment Act move together through Parliament. Bipartisan support on the Labor industry wing, Coalition economic moderates, and crossbench is the governing coalition assembled during 2026–27 consultation.
2029. Operational stand-up. The Dividend Authority is constituted under its enabling legislation and staffs up. The Board of Guardians is appointed. The audit regime is tested with participating providers. The Advisory Panel's first cohort is nominated through public process. Deemed-rate schedule is published for industry comment and finalised before launch.
2030. Launch. First monthly dividend paid. Endowment opens to receive sweep.
This brief is offered to initiate conversation, not conclude it. Specifically, the author seeks:
The full proposal, a technical primer covering the eight domains relevant to defence of the scheme, and the steelman responses to the ten anticipated attacks are available on request.
Correspondence, introductions, and challenges:
michael@australianaidividend.com.au
australianaidividend.com.au
Michael Bois is an Adelaide-based founder and operator. A prior venture, Watt Next, designed a kilowatt-hour-to-credit redemption system structurally analogous to the Dividend. His current operating businesses implement AI in commercial settings directly. The proposal emerges from practitioner engagement with energy metering, merchant integration, digital-first product design, and the economics of customer-facing services.
The proposal is offered in the tradition of citizen-developed policy: an individual with directly relevant experience, proposing a system, inviting institutional development.