A scenario · 2025 – 2032
What being the lucky country costs us, and what making our own luck could look like. One prologue drawn from the record, a fork at a Sydney podium in July 2026, and two tellings of the decade that follows.
listen · narrated by kara bombell
The current trajectory of AI calls for the most ambitious economic reorganisation in Australia's post-war history. Nothing about our circumstances forces us to attempt it. That is precisely the danger. Europe's failure will be loud. Ours will be comfortable, and we will mistake the comfort for success until the decade is over and the decisions have all been made somewhere else. That is one telling. There is another, in which the country notices the comfort in time and decides that a fair go is not something you inherit from luck but something you build, and then share. The prologue below belongs to both. The fork is real, and it happened at a podium in Sydney on 15 July 2026.
This scenario is told through Tess Nguyen, a technology strategy adviser in the Department of Industry, Science and Resources in Canberra, and Jack Halloran, an Australian founder who moved to San Francisco after raising offshore. They are fictional. Until mid 2026, the events around them are not. From August 2026 onward we speculate, and we stop naming real companies: Atlas is the leading American AI firm, Zimo the leading Chinese one. In 2026 Australia has no equivalent to name. Where that fact leads depends on which telling you choose at the fork.
2025
Tess Nguyen works on technology strategy at the Department of Industry, Science and Resources: twenty-nine, four years in, long enough to know how the building metabolises news from abroad. In January, a Chinese lab called DeepSeek releases a model that wipes six hundred billion US dollars off Nvidia overnight. Washington calls it a Sputnik moment. Brussels calls it proof Europe can catch up. Canberra's response is a device policy: DeepSeek is banned from Commonwealth systems within the fortnight, and that is the whole of it. Her oldest friend, Jack Halloran, texts from San Francisco, where he moved after Sydney's funds wouldn't back him. At his dinner tables, artificial general intelligence is discussed like a weather forecast, and nobody has thought about Australia at all.
The year proceeds in the same key. A federal election is fought and won in May without AI rating a question beyond deepfakes. In June, Amazon announces twenty billion dollars of Australian data centres to national celebration, a year after AirTrunk, the local champion, sold to Blackstone for twenty-four billion: the country is becoming the world's most reliable landlord and toasting each new tenancy as if it were the freehold. In August, a productivity roundtable settles the consensus that pleases everyone: no AI Act, light touch, adoption first, we are smarter than Europe. Tess keeps noticing the same flaw. The European error was regulating an industry it didn't have. Australia is congratulating itself on not regulating an industry it doesn't have either.
October compresses the whole national position into one week. In Washington, the Prime Minister signs a critical minerals framework, trading inputs the entire AI race runs on for market access and investment, the things you ask for when you believe you are selling wheat. The same week in Canberra, the Attorney-General rules out the text and data mining exception: copyright will not be weakened for AI, creators will be paid, licensing will be designed instead. It is the most decisive AI policy act Australia takes all decade, and Tess files away the part everyone else misses. The country can act early and hold a line against furious lobbying, but only, so far, in defence of existing value. The instinct exists. It is pointed backwards.
December says the rest. The under-16 social media ban takes effect, world-first, copied abroad within the year. Days earlier, the National AI Plan quietly shelves mandatory guardrails in favour of voluntary guidance, and funds an AI Safety Institute at $29.9 million in the same week the UK announces investments measured in billions. World-leading at restricting the technology, a rounding error on building it. In San Francisco, Jack's company stops hiring graduates: not a freeze, a category deletion. Tess writes a memo arguing that Australia has no position in any of this, no lab, no compute, no stake. It is called a thoughtful contribution, and passed up the line.
january – june 2026
The tent gets more comfortable. A five-year deal puts Microsoft's Copilot across the federal public service, alongside a twenty-five billion dollar framework to expand local AI infrastructure. The government signs a memorandum of understanding with one of the frontier labs and publishes national data centre expectations to shape the wave. And in April, when the most capable model in the world proves so good at finding software vulnerabilities that its maker locks it behind a defensive coalition, Europe is excluded and erupts, while the Australian Signals Directorate, through Five Eyes, is inside the testing perimeter from week one. Canberra's mood is quiet satisfaction. The alliance is paying off.
The cracks are quieter. A seven billion dollar OpenAI data centre planned for Sydney runs into community opposition over energy and water, the first sign that the buildout has a social licence problem the announcements keep skating past. And Tess keeps asking the unwelcome question: what does access cost when you own nothing? An invitation is not an asset. It can be priced, conditioned and withdrawn, and the party issuing it keeps the ledger.
But she has started writing a different memo now, a longer one, the one she has been assembling in pieces for a year. It does not argue that the tent is a trap. It argues that the trust which got Australia inside the tent is itself the asset, the one thing a country of twenty-seven million owns outright, that no amount of American capital can buy, and that the whole world is about to need.
15 july 2026
The Prime Minister takes a podium in Sydney and launches an institution. The Office of AI will sit inside his own department, coordinating new Australian Standards across a remit that runs from energy and copyright through productivity, employment and education to online safety, defence and national security. The standards, he is explicit, will be mandatory: to National Cabinet within the month, to Parliament early next year. He reaches back a century for his analogy, to the arrival of civil aviation, and concedes what this site's prologue has been documenting: that until now the national response has been piecemeal, one issue and one sector at a time. The claim attached to the new office is characteristic: Australia, he says, will be the first country anywhere to pull these questions into a single national framework. World-first, again. The December pattern, again. Unless, this time, it isn't.
The context around the podium tells you why the institution exists, and it is this site's whole argument compressed into one news cycle. The frontier labs, Anthropic among them, have spent the week making their data centre investment conditional on Australia's copyright settings, which is to say a foreign company is publicly pricing a sovereign country's law, and the Prime Minister's answer is that the right guardrails will attract business, with faster approvals and streamlined compliance, while his assistant minister confirms the same morning that no data centre will be buying an exemption from copyright. The October line holds, publicly, under direct commercial pressure. Meanwhile, inside his own party, the former industry minister Ed Husic calls the social-licence approach "doomed to failure", pointing out that self-regulation was tried on big tech for decades and never worked. The comfortable reading and the warning are both in the room, wearing Labor lanyards.
And the substance, when it comes, is startlingly close to this site's second telling, delivered in the government's own voice. The next generation of large data centres will carry a legal obligation to underwrite new power supply, pay their own grid connection and put at least as much energy into the grid as they take out, net-generators funding new renewables and firming, with water rules to match. AI's physical footprint needs Australian land, energy and compute, the Prime Minister argues, which means Australia can set the terms and determine the social licence, if it acts before the building is done. Australia should be more than a data warehouse for products made overseas; dependence means vulnerability; sovereignty is the word he keeps returning to. On copyright he goes further than any leader anywhere: artists retain ownership and control of their work, including its price, and unauthorised training on it gets the plainest word available. "Anything less, is theft."
So the fork, as it actually arrives, is sharper than either telling predicted. Read one way, everything announced is prospective: the legislation is next year, the National Cabinet agreement is next month, and a framework is not a fund. The standards govern where the sheds go and what power they draw; they create no Australian stake in what runs inside them, and there is still not a capability dollar attached. World-first at frameworks, again, the instinct that gave us the under-16 ban now given a letterhead and a timetable. Read the other way, the government has just made the second telling's argument in its own voice, sovereignty, terms-setting, the fair go applied to the machine age, and mandatory obligations on the buildout that no other country has attempted. The words are the second telling's. Whether the decade is depends on what gets appropriated, legislated and owned between this podium and the next election.
This is where the factual record ends. Everything before this podium happened. Everything after it is a wager on which reading the country makes true.
Part two · The scenario
From here, we speculate. Atlas leads the American frontier; Zimo leads the Chinese one. There is no Australian name to introduce, and no chapter in which one appears.
August 2026
Washington formalises what had been informal: a country-tiered licensing regime governing who may buy frontier AI inference, in what volumes, under what reporting. Most of Europe lands in the middle tier and reels. Australia is named Tier 1, alongside the rest of Five Eyes, Japan, Korea and a handful of others.
In Parliament House the announcement is received as vindication. The alliance investment has paid its dividend; the pragmatists were right; somewhere, someone opens champagne. The Australian Standards pass Parliament the following year, mandatory and world-first, and they work: the sheds are net-generators, the water is paid for, the approvals are fast. They govern where the machines sit and what they draw. They create no Australian stake in what runs on them, which is the distinction the decade turns on. Tess writes the memo she has been drafting in her head for a year. Tier 1 access, she argues, is still rationed by price, allocated by a foreign government's discretion, and revocable on ninety days' notice. Australia has been granted a seat at a table it does not own, in a building it cannot enter, under rules it did not write. The memo is called a thoughtful contribution.
2027
A Chinese lab open-sources a frontier-class model, and the offensive cyber capability that had been carefully contained is suddenly everywhere. A ransomware wave breaks across the Tier 2 world. European hospitals and councils pay ransoms in cryptocurrency; their sovereignty policies, which locked them onto second-tier defences, are publicly humiliated. Australia, defended by first-tier American models, is barely scratched. The lesson Canberra absorbs is that dependence works.
Underneath the relief, the real story is starting, and it makes no headlines because it is made entirely of absences. International education, the country's third-largest export, begins its slide: frontier AI tutoring makes a mid-ranked degree look expensive, visa politics does the rest, and enrolments fall for the third consecutive year. Professional services firms discover their billable middle is automatable. Nothing collapses. Things simply stop growing, one sector at a time, and a country whose model is selling services and dirt finds the services half quietly repricing toward zero.
2028
The big four professional services firms halve their graduate intakes in the same quarter, by coincidence of timing rather than coordination. The banks follow. Then, with less fanfare, the Australian Public Service. Continual learning has closed the last gap: the models now build context inside an organisation the way a graduate used to, except in weeks and without a salary band.
Tess's younger brother Minh finished a commerce degree eighteen months ago. They have dinner in Dickson once a month, she pays, and he tells her about application number two hundred and something. He is funny about it, which is worse than if he weren't. The nursing shortage is real, but he faints at the sight of blood. He is not built for the trades. He asks her, because she works in the department with industry in its name, what industry he should aim for. She does not have an answer she believes.
The fiscal arithmetic arrives next, and it is crueller here than almost anywhere, because no comparable country leans harder on personal income tax. The graduate cliff is not just a labour market story; it is a hole in the PAYG base, opening exactly as the NDIS, aged care, defence and interest costs steepen. Resources royalties cannot fill it. The money the automation saves flows to companies headquartered in Delaware, taxed in Singapore, and owned, in part, by Australian super funds whose members are the same people no longer being hired.
2028 – 2029
≈ 87× US advantage
Scenario figures, illustrative. Australian capacity grows; Australian ownership of it does not.
The world is screaming for compute, and on paper Australia should be the answer: the best solar resource in the developed world, empty land, stable law, a trusted flag. The hyperscalers do come. But the transmission queue runs to a decade, the approvals stack across three levels of government, and the connection agreements move at the speed of a system designed for a different century. The capacity that gets built gets built slowly, and every megawatt of it is owned offshore, with profits routed through Singapore and Delaware on their way to somewhere that is not the Australian tax base.
By 2029 the shape of the national position is plain enough that even the optimists have stopped disputing it, they have just stopped mentioning it. Lithium goes out. Rare earths go out. Gas goes out. Land is leased. And intelligence comes back in, by the token, at prices set in San Francisco and adjusted quarterly. Australia has become an input economy for the intelligence age, which is to say it has become what it always was, with better marketing. The difference is that this time the value-add it declined to build is not steel or solar panels. It is the thing that does the thinking.
2029
The Taiwan Strait, which had been deteriorating in increments, deteriorates all at once. Naval confrontations that happened quarterly now happen weekly. Both giants have demonstrated AI-managed weapons platforms in public and better ones in private. And Washington, methodically, calls in two decades of alliance investment: expanded basing in the north, Tindal at full tempo, force posture commitments that were once hypothetical and are now scheduled, and a request, phrased as a question the way such things are, about what Australia will do in the first week of a contingency.
Beijing answers with the other half of the squeeze, and the trade coercion of 2020 returns at scale, with one difference that matters: the iron ore card Australia held last time has gone soft. Simandou is online, Chinese steel demand has peaked, and the leverage that once made coercion survivable has quietly expired. Australia spent twenty years assuming it would never truly have to choose between its security guarantor and its largest customer, and built no position from which to refuse either. The assumption was not tested for so long that it was mistaken for a strategy.
2030
Atlas, now worth more than every listed Australian company combined, goes shopping. It is buying the physical economy: factories, logistics, anything with floor space and grid connection that can be turned to robotics. In Australia the list is short, because the list was always short, and that is its own indictment. The software champion, the one genuinely world-scale product company the ecosystem produced, is acquired in a deal announced as a partnership. The health data assets go next, then the mining automation firms whose autonomous haulage systems were, briefly, the best in the world.
The Foreign Investment Review Board reviews. Conditions are attached. Headquarters will remain in Sydney; a board seat is reserved; the press releases are reassuring. But the licensing rights, the operational control and the upside all run through the acquirer, and everyone signing knows it. The four trillion dollars of superannuation that could have anchored a sovereign position a decade earlier participates in the transaction after all, on the buy side, passively, through index funds that hold Atlas. Australians end up owning a sliver of the company that owns the things they built.
Nobody in Canberra is consulted about the operational decisions that follow. It is not that consultation is refused. It is that it does not occur to anyone to ask.
March 2031
In Washington, the endgame Tess has read about in European cables arrives: an American ultimatum over the last piece of European leverage, a fourteen-hour Council session, a delegation flown across the Atlantic with an ambiguous mandate, six leaders in a room deciding the fate of a continent. Whatever else can be said about Europe, it gets a meeting.
Australia's equivalent happens on a Thursday morning, Canberra time. It is a phone call, secure line, fifteen minutes, from a deputy national security adviser to her Australian counterpart, informing Canberra of new arrangements governing inference allocation, basing tempo and acquisition approvals in the region. The arrangements have been settled between Washington and the companies. The call is a courtesy. There is a pause near the end where a question could go, and the Australian side, having prepared for this call for a week, finds that there is no question that would change anything, and thanks them for the briefing.
Tess is in the corridor outside the secure room when it ends, holding two coffees, one of them for a deputy secretary who no longer needs it. The drama of comfortable vassalage, she thinks, is that there is no room where it happens. There is no fourteen-hour session, no shaking hands gripping a sink, no delegation. There is a calendar invitation that says 0900 BRIEFING (US) and ends four minutes early.
July – August 2032
≈ 106× US advantage
Scenario figures, illustrative. Sovereign-controlled share of the Australian figure: nil.
Brisbane, opening night. The most automated Games ever staged: a driverless transport spine moving two hundred thousand people an hour, a drone perimeter over the river, broadcast orchestrated by AI in forty languages, a ticketing and identity system processing a million credentials a day. For a fortnight, Australia is the most-watched country on earth, and every system doing the watching back is rented.
The attacks begin three weeks out, as everyone knew they would. Probes against the transport orchestration layer. Credential-stuffing waves against ticketing. Something patient and well-resourced testing the stadium's building management systems, the kind of intrusion that is not trying to succeed yet, only to map. Attribution stays deliberately vague in the public statements; privately, the tooling is Zimo-derived, open-weight, and consistent with the grey-zone signalling that has been constant since the Strait crisis of 2029. None of it lands. Atlas's defensive models, operating under the Tier 1 arrangement, hunt and patch in machine time, closing holes faster than they can be found. The Games are flawless. The verdict, worldwide, is that Australia has run the most secure Olympics in history.
Inside the security operations centre at Boondall, the screens are a client dashboard. Escalation authority above a defined threshold sits with Atlas under the licensing terms, and twice during the fortnight the threshold is crossed: decisions about Australian critical infrastructure are made in California in under a second, with Canberra notified in the morning brief. The Australian Signals Directorate, among the best agencies of its kind anywhere, spends the Games watching its own country's marquee event defended by systems it can neither inspect nor direct, billed per token, on terms that renew each September. A minister calls the fortnight a triumph of the alliance. It is. That is the entire problem, stated as a compliment.
Tess watches the closing ceremony in a pub in Karratha, two thousand kilometres from Brisbane and about the same distance from caring who wins the medal tally. The fireworks over the river are magnificent. The pub cheers. Her phone buzzes on the bar.
Part three · Epilogue
Australia's slide was not inevitable. Even in 2026, the country could have changed course. It had every input except the will.
"Australia is a lucky country run mainly by second rate people who share its luck." Donald Horne, The Lucky Country, 1964. He meant it as a warning. It was read as a compliment.
June 2034
Tess lives outside Karratha now, in a house with a view of the solar arrays that power a data precinct she does not own and the autonomous haul roads of a mine that no longer employs drivers. She is thirty-eight. She left the department in late 2031, eight months after the phone call, and took Jack's money, some of it, for a project he insists is the most important of his life: a complete record of how the decade was decided, gathered while the people who lived it can still be asked.
The interviewer on her laptop screen is not a person. He introduces himself as Noah and asks if the appearance is comfortable. She tells him she hates that he isn't real. He says that's fair, and they begin.
You worked under four ministers. Did they understand what was happening?
"They understood pieces. One understood AI was economically important, but important the way the internet was important, a productivity story, a percentage point. Another understood the alliance dimension but thought access was the asset. Nobody put the two together: that we were trading permanent things, minerals, land, energy, market position, for a revocable subscription. The country was never stupid. It was comfortable, and comfort and stupidity produce identical decisions at half the cost."
Take me back to 2026. What did you recommend?
"Six things, across about thirty memos, every one of them called thoughtful. First, compute as the export we refused to invent. We had the cheapest firmed renewable potential in the developed world, stable law, empty land, Five Eyes trust. Special compute zones, ninety-day approvals, transmission built ahead of demand, and one condition Europe never had the standing to impose but we did: capacity hosted here is governed here. Five to ten per cent of global compute under Australian jurisdiction by 2031 was achievable. We built the sheds and let someone else own the machines."
"Second, equity, not adoption subsidies. A sovereign compute vehicle on the Future Fund model, taking direct stakes in infrastructure and labs instead of funding pilot programs for small business chatbots. The capital existed. Four trillion dollars of superannuation sat passively in index funds that ended up financing the acquisition of Australian assets. A mandated infrastructure allocation was politically hard and structurally trivial. We chose neither."
"Third, minerals for technology, not tariff relief. The 2025 framework traded our one genuine chokepoint for market access, the thing you request when you believe you are selling commodities. The alternative was conditionality: rare earths and lithium flow in exchange for guaranteed inference allocations, crisis-time model access, co-located processing and a stake in what the minerals become. We held inputs the entire race needed and priced them like wheat."
"Fourth, the middle-power table. Japan, Korea, Singapore, Canada, the Netherlands. Same position as us, between two empires that valued them instrumentally, each holding a piece of the supply chain. Australia chaired everything in the region except the conversation that mattered. We never sat down together and said: we have more leverage jointly than severally, let's price it that way."
"Fifth, flexicurity on the HECS chassis. An income-contingent retraining entitlement, paired with wage insurance, administered by the tax office that already ran the mechanism. The instrument was forty years old. The graduate cliff could have been a transition. Nobody repurposed it, so it was a cohort write-off instead, and my brother was in the cohort."
"Sixth, the one nobody remembers we were positioned for: safety and assurance as the national niche. A country of twenty-seven million was never going to out-build the labs, but it could have out-audited them. We had a signals directorate trusted across every alliance, a regulatory culture other countries actually believed, and a safety institute funded at $29.9 million, which is to say, funded as a gesture. Scale that a hundredfold into a national evaluation and assurance authority: red-teaming, certification, incident response, independent model audits, sold as services to every middle power that trusted neither Washington nor Beijing to mark its own homework. By 2032 the world was desperate to buy exactly that capability, and bought it from the companies that needed auditing. We could have been the country whose stamp on a system meant something. It was the cheapest of the six, and the only one with no natural enemy in the building. It died of indifference rather than opposition, which somehow makes it worse."
The Brisbane Games were flawless. Doesn't that vindicate the strategy?
"It vindicates the product. The Games were secure the way a tenant's house is dry: genuinely, and on someone else's terms. Twice in that fortnight, decisions about Australian critical infrastructure were taken offshore in under a second, and we read about them in the morning brief. The most-watched fortnight in our history, defended flawlessly, by the token, under a contract that renewed the following month. Brisbane should have been the showcase for an Australian assurance industry, our systems certifying and defending our event, the stamp the rest of the world came to buy. Instead it was the world's most expensive product demonstration, and we were the venue. If that's vindication, the word has stopped meaning anything."
Why did none of it happen?
"Because nothing forced it. That's the honest answer. Europe got punished in public, ransoms, downgrades, riots, and even Europe couldn't act in time. We were never punished at all. Every quarter, the terms of trade held, the access continued, the sheds went up, and anyone arguing for sacrifice was arguing against the evidence of everyone's eyes. You cannot scare a country with a forecast while the surf is good. Horne told us in 1964 that the luck was the problem, that it let second-rate decisions look first-rate for decades at a time. We put the phrase on tea towels."
Are you angry?
She is quiet for a moment. Through the window, the late sun catches the solar arrays, and beyond them a road train of autonomous haulers moves north, perfectly spaced, toward a port that loads ships bound for the factories of other countries.
"I used to think the luck was the minerals. The sunshine. The distance from everything. It wasn't. The luck was time. We had a full decade of warning, longer than Europe, longer than almost anyone, delivered in plain language by the people building the thing. Other countries spent that decade. We banked it. And the thing about banked time is that it pays no interest and it cannot be withdrawn. So yes. I'm angry. Not at any minister. At the comfort. At how good the comfort felt, the whole way down."
That was the warning.
part two · the other telling
From here, we speculate. Atlas leads the American frontier; Zimo leads the Chinese one. In 2026 there is still no Australian name to introduce. By the end of this telling there are several, and more to the point, there is a stake: owned widely, governed here, and built to be shared.
august 2026
Washington formalises what had been informal: a country-tiered licensing regime governing who may buy frontier AI inference, in what volumes, under what reporting. Most of Europe lands in the middle tier and reels. Australia is named Tier 1, alongside the rest of Five Eyes, Japan, Korea and a handful of others.
In Parliament House the announcement is received, at first, exactly as the warning predicted: as vindication. The alliance investment has paid its dividend; the pragmatists were right; somewhere, someone opens champagne. Tess sends her memo up the line, the long one, and braces for the usual verdict.
This time it is not the usual verdict. The memo reaches a Treasurer who has spent the winter reading the European cables and has arrived at a private conviction he is not yet saying in public: that Tier 1 access is a subscription, that subscriptions get cancelled, and that the only access worth having is the kind you cannot have revoked because you helped build the thing. The memo's argument is simple enough to survive a Cabinet submission. Australia should accept Tier 1 with both hands, and treat every month of it as a bridge to a position the country actually owns. Access buys time. The question is what you build with the time.
What lands in the October budget is not a slogan. It is a vehicle. A sovereign compute and capability fund, modelled on the Future Fund, seeded from the resources rent and mandated to take direct equity, not grants, not pilot programs, but ownership stakes in the infrastructure, the labs and the assurance firms that will sit on Australian soil. And a quiet, consequential clause that the super funds fight and then accept: a sovereign infrastructure allocation, so that the four trillion dollars of Australians' retirement savings stops passively financing the buildout from the buy side of someone else's balance sheet, and starts owning a slice of the buildout that happens here. The principle is stated in one sentence the Treasurer repeats until it sticks: if it is built in Australia, Australians own a real piece of it.
late 2026 – 2027
The thing about the graduate cliff is that everyone can see it coming, which is exactly why, in the warning, nobody moves until it arrives. Australia does the opposite. Before the layoffs start, it builds the institution to catch them.
The Flourish Commission is established by statute in the last sitting fortnight of 2026, and the design is deliberate. The Productivity Commission keeps the economic half of the transition: how the gains are produced. The Flourish Commission takes the human half: how they are shared, and how nobody is converted from a person with a future into a line item in a cost saving. It is not a welfare body. It is a transition authority, and it has teeth. Firms that automate are not prevented from automating, but to claim the productivity incentives they must run their workforce reductions through the Commission's scheme: redeploy, reskill, or fund the bridge for those they release. Australia gives the manoeuvre a name so the country can argue about it honestly, the AI Layoff Trap, the false economy of dumping a trained cohort onto the street and discovering eighteen months later that you have hollowed out your own customers and your own tax base at once.
The chassis is forty years old and already built: the income-contingent loan system that put a generation through university becomes the rail for an income-contingent retraining entitlement, paired with wage insurance, administered by the tax office that already runs the mechanism. What is new is who decides. The Commission governs through participatory mechanisms, citizens' panels, sector assemblies, regional boards, on the principle that a transition this large cannot be designed by the people it is not happening to. The fair go, written down at last not as a sentiment but as a procedure: those affected hold seats at the table where the decisions are made.
Tess's younger brother Minh finished a commerce degree eighteen months ago and has sent out two hundred and something applications. In the warning, he is a cohort write-off, and Tess has no answer she believes. In this version, the Commission routes him into the one sector that is hiring faster than it can train, and that Australia has decided to make its own: assurance. He starts a paid traineeship in model evaluation. He is funny about it, the way he was funny about the rejections, except now the joke has a different ending.
2027
A Chinese lab open-sources a frontier-class model, and the offensive cyber capability that had been carefully contained is suddenly everywhere. A ransomware wave breaks across the Tier 2 world. European hospitals and councils pay ransoms in cryptocurrency. Australia, defended by first-tier American models, is barely scratched, and in the warning, that is the whole lesson, and it is the wrong one: dependence works, until the day it is reviewed.
Australia draws the other lesson. Surviving on someone else's immune system is not health, and the way you turn borrowed defence into a durable position is to become the country that certifies whether anyone's defence, anyone's model, anyone's AI system, is actually safe and actually fair. The $29.9 million AI Safety Institute, the rounding error of the warning, is scaled a hundredfold into a national evaluation and assurance authority. The Office of Artificial Intelligence, launched in July 2026 to coordinate standards across energy, copyright and labour, turns out to be the chassis: the standards it wrote become the tests the authority administers. It has the one input that cannot be bought: the Australian Signals Directorate's standing across every alliance, and a regulatory culture that other countries, for reasons of history and temperament, actually believe.
But Australia makes a bet that goes one step past the obvious. The world is filling up with technical safety labs that will tell you whether a model can be jailbroken or made to leak. Australia decides to certify something harder and rarer: not just whether a system is secure, but whether it is ethical. Does it distribute its harms and its benefits fairly? Who carries the downside? Does it serve the flourishing of the people it acts on, or only the margins of the people who own it? Ethical assurance becomes the distinctively Australian product, the stamp that asks the questions the labs marking their own homework will never ask themselves. It turns out to be exactly what the moment wants. The global backlash against an extractive AI elite is cresting, and an Australian certification is the thing both kinds of government quietly need: the autocracies want it because a system audited for fairness is a system less likely to trigger the unrest that threatens regime stability, and the democracies want it because their citizens have started demanding that someone independent, trusted, and not on the payroll, vouch for the machines now running their lives.
Minh, six months in, sits his first certification exam. He passes. His name goes on a register that, within three years, the world will be checking.
2027 – 2029
The world is screaming for compute, and on paper Australia is the answer: the best solar resource in the developed world, "empty" land, stable law, a trusted flag. In the warning, the hyperscalers come anyway, but the transmission queue runs to a decade, the approvals stack across three levels of government, every megawatt is owned offshore, and the value drains out through Singapore and Delaware. The dirt and the sun go out; the tokens come back in, priced in San Francisco.
Australia decides to keep the value where the sun falls. It does not do this by going faster and cheaper than everyone else, the race it would lose. It does it by attaching conditions it is uniquely positioned to impose, and discovering that the conditions are not a deterrent but the product.
Infrastructure sovereignty becomes law in stages, and it starts from the floor the fork laid down: the 2026 obligations that made data centres net-generators paying their own way are extended upward, from how the sheds are powered to who owns what runs inside them. Hosting capacity built in Australia must be governed under Australian jurisdiction, the clause Europe never had the standing to demand and Australia, holding the sunlight and the minerals, does. Builds must meet a hard environmental standard, powered by firmed renewables, because the same gas-tax-and-sovereign-wealth settlement that funds the transition also funds the cleanest grid in the developed world, and a clean grid turns out to be a selling point: by 2029 the multinationals competing for AI capacity are under their own pressure to show clean power, and Australia has the cleanest on offer. Construction phases run under union agreements, so the buildout produces secure work, not just contracts. And every project carries a local ownership and control requirement, the sovereign fund or the super allocation holding a real stake, so that the machines on Australian soil are, in part, Australian-owned.
And there is a deeper condition, the one that ends up mattering most, because it changes whose sovereignty is being won. The land the warning kept calling empty was never empty. The compute zones, the transmission corridors and the new haul roads all run across Country that has been stewarded for sixty thousand years, and this time Australia builds with its custodians rather than around them. The regulations that define how and where data infrastructure can go are written with Aboriginal and Torres Strait Islander custodians holding genuine decision-making rights, not consulted once the route is already drawn but deciding, alongside the country, what happens on their Country. And the benefit-sharing is structural, not a plaque and a Welcome: traditional owners hold equity in the precincts built on their land, seats on the bodies that govern them, and a standing share of the revenue the sovereignty throws off. It is, in plain terms, much of what the 2023 referendum asked the country to enshrine and the country declined, arriving by another road, through the infrastructure rather than the Constitution. And underneath it is a quieter recognition: the posture Australia spent the whole decade reaching for, custodianship instead of extraction, value kept in place and shared rather than dug up and shipped out, was one its First Peoples had been practising for longer than there had been an Australia to learn it from. The country stops filing that under heritage and starts treating it as the model.
The hard part is honest. The gas industry fights the resource-rent reform with everything it has, and for a while it is winning. The 2028 fight over the gas tax is the ugliest domestic politics of the decade, an industry campaign, a wobble in the polls, a Senate that nearly folds. It passes by a margin, late one night, because a Treasurer is willing to stake a government on the argument that a country sitting on this much sunlight and this much gas has no excuse to let the proceeds leave, and because enough Australians, asked plainly, decide they agree. It is not clean and it is not easy. It is, in the precise old sense of the phrase, hard won.
By 2029 the shape of the national position is plain, and for once it is the optimists who are right and saying so. Lithium goes out, and a stake in the processing comes back. Rare earths go out, and a guaranteed compute allocation comes back. The sun is converted, on Australian soil, under Australian law, by Australian-owned firms in part, into the one thing the warning said the country would only ever buy by the token: intelligence, generated here. Australia is becoming a place that adds value to its own inputs, which it had been told for a century it could never afford to do. The thing it declined to build in the warning, the thing that does the thinking, it builds.
2029
≈ 33× and closing
Scenario figures, illustrative. Australian-jurisdiction capacity with sovereign fund, super and traditional-owner equity stakes: 41% and rising.
The Taiwan Strait, which had been deteriorating in increments, deteriorates faster. Both giants have demonstrated AI-managed weapons platforms. And Washington, methodically, begins to call in two decades of alliance investment: basing, posture, a request phrased as a question about what Australia will do in the first week of a contingency. Beijing answers with the other half of the squeeze, and the trade coercion of 2020 returns at scale.
In the warning, this is where the bill arrives and Australia cannot pay it, because it built no position from which to refuse either giant. In this version, Australia is not standing alone in the doorway. It is sitting at a table it spent three years building.
The middle-power coalition is the least glamorous and most important thing Australia does all decade. Japan, Korea, Singapore, Canada, the Netherlands: countries in exactly Australia's position, valued instrumentally by two empires, each holding a different piece of the supply chain, each tired of being asked to choose. Australia, holding the minerals and increasingly the clean compute and the assurance standard, convenes them, and the pitch is the one Tess wrote years earlier and nobody read: we have more leverage jointly than severally, so let us price it that way. The Netherlands holds the lithography. Japan and Korea hold the fabrication and the materials. Australia holds the inputs, the hosting, and the stamp that certifies whether any of it is safe and fair. Priced together, it is not a plea. It is a position.
It does not stop the Strait from being dangerous. Nothing a middle power does stops that. What it does is change what Australia is for. When Washington asks and Beijing threatens, Australia is no longer a supplicant offering loyalty in exchange for continued access. It is a member of a bloc that supplies things both giants need, governed by a standard both giants are starting to have to meet, and able to say, for the first time in the decade, that its answer will be its own.
2030
Atlas, now worth more than every listed Australian company combined, goes shopping. It is buying the physical economy everywhere it can: factories, logistics, health data, mining automation, anything with floor space and grid connection and upside. In the warning, the Australian list is short because the list was always short, and what little exists is acquired in deals announced as partnerships, with the licensing rights and the control and the upside all running offshore, and the country's super savings ending up, passively, on the buy side of its own dispossession.
In this version the list is longer, because Australia spent four years building things worth buying, and the things are not for sale on those terms. The local ownership and control requirements that looked like red tape in 2027 are, in 2030, a moat. When Atlas comes for the autonomous-haulage firms, the health-data assets, the assurance companies, the Foreign Investment Review Board does not merely attach reassuring conditions to a foregone conclusion. It can say no, and mean it, because the sovereign fund and the super allocation already hold blocking stakes, and because the capability is governed under Australian law that does not transfer with the cap table.
The clearest case is the one that started in San Francisco. Jack's company, the one genuinely world-scale product the Australian-born ecosystem produced, gets the Atlas approach: a number with a comma in a new place, a deal that would be announced as a partnership. In the warning, he is in no position to refuse, and the company becomes a line in someone else's annual report. In this version, the sovereign compute fund and a syndicate of the super funds offer the other thing a founder needs, patient capital at scale, anchored to a jurisdiction that now has compute, a coalition, and a reason to keep the company whole. He takes the Australian money. The second time in his life he chooses an address, he chooses Sydney.
He comes home. Not as a gesture, and not at a loss. He comes home because, for the first time since he left, home is where the better deal is.
march 2031
In the warning, Australia's place in the endgame is a phone call: a secure line, fifteen minutes, a deputy national security adviser informing Canberra of arrangements already settled between Washington and the companies, a pause near the end where a question could go and there is no question that would change anything. Comfortable vassalage, with no room where it happens.
This version has a room. It is not a grand one, and Australia is not the largest country in it, and the meeting does not go entirely Australia's way. But when the arrangements governing inference allocation and regional posture are settled, Australia is at the table, not on the receiving end of a courtesy. It is there because it brought things to the table that have to be allocated: clean compute under its own jurisdiction, a coalition that supplies what both giants need, a minerals position priced as leverage, and an assurance standard that has quietly become the thing a system has to pass before allies will field it. The deputy who once would have made the call now sits across from a counterpart, because you cannot brief a partner whose signature you require.
Tess is in the room, no longer carrying coffee for someone who outranks her. She has spent the decade being told her memos were thoughtful contributions. The arrangements that get signed carry, in three places, language she drafted. The drama of making your own luck, she thinks, is that it is even less cinematic than losing. There is no fourteen-hour session and no delegation flown across an ocean with an ambiguous mandate. There is a meeting that starts on time, runs long, and ends with Australia's name on the document as a party, not a recipient.
july – august 2032
≈ 23× and closing
Scenario figures, illustrative. Majority of Australian capacity under Australian jurisdiction, with sovereign, super and traditional-owner stakes. Coalition pooled capacity: larger than either number alone.
Brisbane, opening night. The most automated Games ever staged, and the most watched: a driverless transport spine, a drone perimeter over the river, broadcast orchestrated in forty languages, a million credentials a day. In the warning, every system doing the watching is rented, the security operations centre is a client dashboard, escalation authority above a threshold sits in California, and twice during the fortnight decisions about Australian critical infrastructure are made offshore in under a second. The Games are flawless and the country is a venue.
This time the country is a host in the older sense of the word. The attacks come, as everyone knew they would, the same probes, the same credential-stuffing, the same patient grey-zone mapping, the tooling Zimo-derived and consistent with the signalling that has been constant since the Strait crisis. They do not land. But the systems that hunt and patch them are not only rented. They run on Australian compute, under Australian jurisdiction, certified to the Australian assurance standard, operated by Australians, with the coalition's shared defence behind them and the escalation authority held, this time, in Boondall, not in California. Inside the security operations centre, the screens are not a client's dashboard. They are the country's. Minh is on shift for two of the fourteen nights. The brother who could not get an interview in the warning is, in this version, one of the people keeping a country's marquee event safe.
And Australia does the thing the fair go was always supposed to mean: it does not keep the capability to itself. The assurance methods, the evaluation tooling, the certification framework, are shared with the coalition and offered to the Global Majority on terms designed for adoption, not extraction, so that the smaller countries arriving in Brisbane are defended by the same standard, and leave able to run a version of it at home. The Games are not pitched as a triumph of one nation's technology. They are pitched, and run, as a demonstration that the safe and fair deployment of AI is something a middle power can lead and then hand around.
There is one more thing about Brisbane 2032, and it is not a technical thing. The rising tide of right-wing extremism that ran through the late 2020s, in Australia as everywhere, had to be turned before the country could credibly invite the world, and turning it was its own decade-long fight, fought in the same spirit as the rest: not luck, but work, and the deliberate choice that prosperity built on a fair go has to be a prosperity that includes everyone inside the borders before it can mean anything to those arriving from outside them. By 2032 the choice has held. Every athlete, every delegation, every visitor who lands for the Games knows, because it has been made true and not merely promised, that they will be welcomed and protected here, whoever they are. The most-watched fortnight in the country's history is also, quietly, the proof of what kind of country was watching back.
part three · epilogue
Australia's recovery was not inevitable either. Even in 2026, with every input in hand, the country could have chosen the comfortable slide, and very nearly did. What follows is the other ending, told by someone who spent the decade making it.
"Australia is a lucky country run mainly by second rate people who share its luck." Donald Horne, The Lucky Country, 1964. He meant it as a warning. For sixty years it was read as a compliment. Then, once, it was read correctly.
june 2034
Tess lives outside Karratha now, in a house with a view of the solar arrays that power a data precinct the sovereign fund part-owns, and the autonomous haul roads of a mine that retrained its drivers rather than discarding them. She is thirty-eight. She left the department in late 2032, not in defeat but because the work had moved: she chairs an assurance body now, and she took some of Jack's money, and some of the country's, for a project they insist is the most important either of them will do, a complete record of how the decade was decided, gathered while the people who lived it can still be asked.
The interviewer on her laptop screen is not a person. He introduces himself as Noah and asks if the appearance is comfortable. She tells him she still hates that he isn't real. He says that's fair, and they begin.
You worked under four ministers. When did they understand what was happening?
"Pieces, at first, the way the warning describes. One understood AI was economically important, the way the internet was important, a productivity story. Another understood the alliance dimension but thought access was the asset. The thing that changed wasn't that someone finally understood all of it. It was that one Treasurer understood enough, at the right moment, to act before the evidence forced him to, which is the only kind of acting that counts. The country was never stupid. It was comfortable. The whole achievement of the decade was deciding, while the surf was still good, that comfort was no longer the same thing as success."
Take me back to 2026. What did the turn actually consist of?
"Six things. People talk about them now as a strategy, as if they arrived as a set. They didn't. They arrived as about thirty memos, every one of them once called thoughtful, and then, for the first time, acted on."
"First, compute as the export we'd refused to invent. We had the cheapest firmed renewable potential in the developed world, stable law, land, Five Eyes trust. So we built special compute zones, approvals measured in months, transmission ahead of demand, and the one condition Europe never had the standing to impose and we did: capacity hosted here is governed here. We stopped building sheds for other people to own the machines inside. We built the machines, and kept a stake in them."
"Second, equity instead of adoption subsidies. A sovereign compute and capability fund on the Future Fund model, taking direct stakes in infrastructure and labs, not handing out chatbot vouchers to small business. And the part nobody thought we'd get, a mandated infrastructure allocation for superannuation, so that four trillion dollars of Australians' own savings stopped passively financing the buyout of Australian assets and started owning the buildout that happened here. The capital existed the whole time. We finally pointed it at ourselves."
"Third, minerals for technology, not tariff relief. The 2025 framework had priced our one genuine chokepoint like wheat. We renegotiated the posture, not overnight, and not alone: rare earths and lithium flow, and in exchange we get guaranteed inference allocations, crisis-time model access, co-located processing, a stake in what the minerals become. We held inputs the whole race needed, and the second time we priced them like it."
"Fourth, the middle-power table. Japan, Korea, Singapore, Canada, the Netherlands. Same position as us, between two empires that valued us instrumentally, each holding a piece of the supply chain. We'd chaired everything in the region except the conversation that mattered, so we finally called it: we have more leverage jointly than severally, let's price it that way. When the Strait crisis came, we weren't standing in the doorway alone."
"Fifth, the Flourish Commission, which is the one I'd save if I could only save one. Flexicurity on the HECS chassis, an income-contingent retraining entitlement paired with wage insurance, run through the tax office that already had the rails. But more than the mechanism, the principle: a transition authority that governs how the gains are shared, sitting alongside the Productivity Commission that governs how they're made, with the people affected holding seats at the table through participatory governance, not consulted after the fact. The graduate cliff became a bridge. My brother walked across it. And the gains we redistributed weren't abstract, they fully funded the NDIS, they held up the care economy, they made the fair go a line in a budget instead of a line in a speech."
"Sixth, the one we were uniquely built for: assurance as the national niche, and then a step past it. A country of twenty-seven million was never going to out-build the labs. But we could out-trust them. We scaled the safety institute a hundredfold into a national evaluation and assurance authority, ASD's standing behind it, our regulatory credibility in front of it. And we didn't stop at technical safety. We certified whether systems were ethical, whether they distributed harm fairly, whether they served the people they acted on or just the people who owned them. The world was desperate for exactly that, and tired of the companies marking their own homework. We became the country whose stamp meant something, morally, not just technically. By Brisbane, the world was buying it from us."
There is a decision that doesn't appear in your six. The way the build-out happened on Country, with First Nations custodians.
"It doesn't appear as a seventh thing because it wasn't a separate thing. It ran through all six. For two centuries we'd treated reconciliation as something you do with words, a statement, an apology, a referendum question, and every time the words arrived the material reality stayed exactly where it was. The infrastructure gave us a different instrument. When the compute zones and the transmission corridors and the precincts went onto Country, traditional owners weren't consulted at the end of the process, they held decision rights at the start of it and equity for the life of it. We didn't ask the nation to feel differently about its past. We changed who owned the future being built on top of it. The 2023 referendum had failed, and a lot of people read that as the country saying no to reconciliation. It wasn't. It was the country saying no to a particular form of words. When we offered a form of ownership instead, the same country said yes, project by project, because a fair go turns out to be easier to grant in a stake than in a sentence. By the Games, the most advanced infrastructure we had sat on Country that shared in it, governed in part by the people who had cared for it longest. That was the reconciliation, the real one, the kind that shows up on a balance sheet. We stopped performing sorrow about the past and started co-owning what came next. I think it's the thing I'm proudest of, and it's the thing the warning never imagined we'd reach for, because in the warning we never built anything worth sharing in the first place."
The Brisbane Games were flawless. In the other telling, that flawlessness is the indictment. Here it's the proof. What's the difference?
"Ownership, and generosity. In the warning, the Games are secure the way a tenant's house is dry, genuinely, and on someone else's terms, with the real decisions taken offshore in under a second. In what we built, the systems ran on our compute, under our jurisdiction, certified to our standard, operated by our people, with the escalation authority held in Boondall. And then we gave the capability away, to the coalition, to the Global Majority, on terms built for adoption instead of extraction. The most-watched fortnight in our history wasn't a product demonstration with us as the venue. It was a demonstration that a middle power can make the deployment of this technology safe and fair, and then hand the method around. That's the whole difference. Not that the Games were secure. That we were the ones securing them, and we didn't keep it for ourselves."
Why did it happen this time, when the warning says it so easily might not have?
"Because we stopped waiting to be forced. That's the honest answer, and it's almost embarrassing how simple it is. Europe got punished in public, and even Europe couldn't act in time, because by then there was nothing left to act with. We were never punished, which in the warning is exactly why we never moved. The thing we did differently was refuse to treat the absence of a crisis as permission to do nothing. You cannot scare a country with a forecast while the surf is good. So we didn't try to scare it. We made the argument that a fair go is something you build and share, not something you inherit and spend, and enough people, asked plainly, agreed. Horne told us in 1964 that the luck was the problem. We finally read it as the warning it was, and then we did the thing the second half of his sentence describes, almost by accident. We made luck worth sharing."
Are you proud?
She takes a moment. Through the window, the late sun catches the solar arrays, and beyond them a road train of autonomous haulers moves north, perfectly spaced, toward a port that loads ships still bound for other countries, but lighter now, because more of the value leaves as something we made rather than something we dug.
"I used to think the luck was the minerals. The sunshine. The distance. It wasn't. The luck was time, a full decade of warning, longer than almost anyone got, delivered in plain language by the people building the thing. The warning says other countries spent that decade and we banked it. We didn't bank it. We spent it, on the dullest, least heroic things in the world, committees and allocations and union agreements and a gas tax that passed by two votes at one in the morning. None of it was luck. Every bit of it was a choice somebody had to win. So, proud? Of the country, yes. But mostly I'm relieved, because I know exactly how close we came to the other ending, and how good the comfort felt the whole way up to the edge of it. We didn't escape because we were special. We escaped because, this once, we decided not to be lucky. We decided to be good at it instead."
That was the choice.