Model access is now a services import of five to eight billion dollars a year, growing twenty to twenty five per cent a quarter. Those are the Commonwealth's own numbers. Nobody had compounded them.
Even on the lowest base and the slowest rate, this passes Australia's iron ore export earnings inside five years and exceeds total Australian imports by year seven. Which cannot happen. So the curve bends. The only question is what bends it.
Cost for each unit of capability has been falling throughout. Cheaper inference has raised total spending, not lowered it.
Demand is nowhere near it, and nothing in the adoption curve suggests it soon will be.
Australia substitutes domestically, or another government decides for us. The second already happened once, in June 2026. Access came back as a grant, not a right.
The largest single share of every dollar leaving this country is not hardware and it is not energy. It is people.
On the Commonwealth's own value chain figures. Model development, research, tuning, evaluation, integration, and the institutional work of putting any of it to use.
Australia will not build a frontier model and I do not think it should try. What is buildable is smaller. Built on open foundations, tuned on proprietary data, operated under Australian control, inside the sectors that cannot tolerate the alternative.
The data cannot leave the institution, and the regulator is already in the room.
The operating knowledge is the asset. Social licence decides what gets built.
The sovereignty requirement answers itself.
Consent is the binding constraint and the data is the most sensitive the country holds.
In every case the model is the modest part. The apparatus around it is the difficulty, it is sector specific, and it is the fifth pillar.
Government, industry and research each hold part of what sovereign capability requires. Each has a sound reason not to move first. I arrange the first move as a single move.
Set it, then test it against the parties who would have to act.
Commission the analysis a decision has to survive.
Structure the funding that converts intent into a programme.
Governance, partnerships, team. Then hand it over.
I build no models, operate no compute and take no success fees. There is nothing I am selling into the decisions I am helping to shape.
I am Steve Brittan. My background is architecture and systems design, design thinking as a method for institutional change, and executive leadership, most recently as President and Chief Executive Officer of a college in California. Before that, principal and managing partner of an international architecture and engineering practice, acquired by a public company. I have held teaching appointments at Columbia University, Harvard University and the Rhode Island School of Design.
Data, compute, model, policy. Sound as far as it goes, and incomplete in a way that changes how instruments get designed.
People aren't failing to think. They're trying to think in environments that are engineered to stop them from thinking.
Attention is fragmented deliberately, because fragmented intelligence is easier to monetise. A mind in that condition cannot hold a long argument, navigate real complexity, or picture something that does not yet exist. Which is precisely what sovereign capability asks of people.
Hold that against a university. The fragmented mind Hansen describes is the student body. It is also the faculty, the executive and the board. These are the institutions being asked to absorb the most demanding technology of the century, and very little in how they are currently designed protects the conditions under which real thought can happen.
Designing institutions that protect thought, culture, memory and judgement is not a cultural nicety running alongside the technical work. It is the condition on which the technical work depends, and it is the fifth pillar.
The old systems are in decline. That is the same thing as saying they are available to be redesigned.
Held in Australia, governed under Australian law.
In country capacity under Australian control.
Capability no other jurisdiction can revoke.
Our own rules, on our own timetable.
The capacity to turn the other four into working capability.
The four pillar framework is the work of Hyland-Wood, Perrin and Guihot at the Queensland University of Technology. The fifth is mine.
Policy treats talent as a product of infrastructure. Build the compute, and the people follow. They do not.
It is people who know what to do with them, working inside institutions able to absorb the work.
An instrument built on the wrong sequence underdelivers several years after the capital is committed, when the framing has hardened.
People and practice is not a humanistic addition to a technical framework.
On the Commonwealth's own value chain figures, the largest single share of every dollar Australia spends on model access is people. The argument arrives from the infrastructure side, not the other one.
The fifth pillar is the argument. Building it inside an institution is the engagement.
Government, industry and research each hold part of what sovereign capability requires. Each has a sound reason not to move ahead of the other two. I arrange the first move as a single move.
Holds the objective and the instruments. Lacks an independent evidence base on the demand side.
Holds the data, the operating problem and the capital. Lacks a defensible position on governance, assurance and consent.
Holds the disciplinary depth and the standing to hold a public standard. Lacks a route into the rooms where it would be used.
A proposition specific enough to be signed rather than admired, and an honest account of who is prepared to sign it.
The analysis a decision has to survive. Demand modelling, assurance method, consent evidence that holds under appeal rather than in a brochure.
Public instruments and institutional capital. At infrastructure scale in this country that means superannuation, not venture networks.
Governance, partnerships, workforce architecture, and the standing team that runs it once I step back.
The first phase is severable by design. Stop at the gate and you keep everything produced up to it.
Nought to ninety days. Position tested, instruments and counterparties mapped, a signable proposition. Formal decision point closes the phase.
Months three to twelve. Agreements executed, funding secured, research under way, first engagements invoiced by the client rather than by me.
From month twelve. Permanent unit, named director, operating team. Method, contacts and work product transfer in full.
Not a model developer, a systems integrator, an infrastructure operator or a lobbyist. No panel arrangement, and I do not bid for delivery work created by my own advice.
Every party can talk to me without calculating what I am selling them.
A conflict of interest register lodged before any engagement begins, and maintained throughout.
From any procurement or selection decision involving a registered party.
The client is the only source of payment for the duration.
Evidence that is paid by the result is not evidence.
Method, work product and contacts, at any point, including at the gate.
Where things are sited and who lives there. How an approval is assembled and what evidence it will bear. Who inside an institution has standing to act. What a community is actually being asked to consent to.
These are design and planning problems with an established literature and established methods. They are not being treated as a discipline, which is why they keep arriving late in a programme, when they are most expensive to fix.
What is the twenty first century social contract?
Turning technological progress into social progress has always taken deliberate institutional effort. Last time the settlement was assembled retrospectively, over a century. Australia is building this stack deliberately, which raises the question of whether the settlement can be designed alongside it.
The path was precipitous well before artificial intelligence arrived. What the disruption does is force the question of the social contract open again, while there is still time to answer it deliberately. I do not expect the opportunity to come twice.