The Economics of the AI Transition

Aschenbrenner is right about the trajectory.¹ AGI by the late twenties on the scaling curves, an intelligence explosion that compresses a decade of research into one, trillions of dollars and tens of gigawatts deployed against it. Situational Awareness is the cleanest statement of the road we are on, and the chapters that follow accept it as the road. What this essay is about is what happens to the economy as we walk it, and what happens to the welfare state that runs on the income-tax base after the income-tax base stops growing.

The mechanism is a curve gap. Software AI raises productivity on the inference-cost curve, which falls roughly fifty-fold per year. Robotics raises productivity on the manufacturing-capacity curve, which scales by orders of magnitude per decade. The interval between them — software-cheap, robots-still-people — is the Baumol window: the cognitive layer of every service sector is being unbundled while the physical layer is still bodies in rooms. Inside that window, public-sector wages chase a private-sector wage curve that has just steepened, and the services those wages buy do not. The labour share of national income falls. The rent flows to a small set of firms whose capital is non-rival, partially excludable, and concentrated at the substrate of the AI economy. The welfare state, whose tax base is wages, cannot follow the rent if the rent does not flow through wages.

Denmark is the cleanest case. Public-sector employment is thirty per cent of the workforce against an OECD average of twenty-one. The services the state buys are exactly the ones that take a body in a room — eldercare, nursing, primary education, courtrooms. The indexation rule that ties public-sector pay to private-sector pay (the reguleringsordningen) runs at a two-year lag, which means the wage spike from chapter III arrives in the budget after it has already arrived in the market. The fiscal-sustainability indicator was already drifting down before the Baumol-window wage curve had finished arriving. The institutional architecture that could capture the rent and redistribute it — ATP at roughly DKK 925 billion, the doctrinal precedent of folkepension as universal-residence redistribution — exists. The political technology to point that architecture at AI rent does not.

The economic argument runs to five on-record predictions. The series falsifies on each.

The Baumol window closes when humanoid manufacturing catches up. Goldman Sachs projects 1.4 million units per year by 2035; Citi projects 13 million. My claim is under one million annual global humanoid shipments before 2030. Verified deliveries crossing one million annually before then falsify the chapter that depends on the gap.
US public-sector real unit-labour-cost does not fall year-on-year before 2028. The wage curve is moving faster than the public-sector services it buys can substitute. A year-on-year decline before 2028 falsifies the mechanism.
Danish private-non-farm wage share falls by at least three percentage points by 2032. The compression is structural to the rent-capture argument and the Baumol-window wage spike together. Failure to decline at that rate over the period falsifies the labour-side reading of the series.
The Danish 10-year statsobligation trades at a sustained positive spread to the German Bund by 2028. The spread turned negative during 2025; my claim is that the negative episode is the bottom of the range, and that fiscal pressure from the eroding labour-income tax base reprices it positive within three years. A sustained negative or zero spread through 2028 falsifies.
Open-weight frontier-capability lag holds at roughly three months or widens through 2028. A compression below six weeks on the Epoch Capabilities Index, or the open-weight frontier overtaking the closed frontier outright, falsifies the substrate-rent-durable assumption and rewrites chapters V through VII.

These are the operator’s specific bets. The chapters add more where the evidence pulls them out. The accountability ledger lives at essays/PREDICTIONS.md.

The companion piece to read alongside is No Winning Strategy (May 2026). That essay does the substrate-lockdown exterior — why countries outside the AI production chain cannot catch up, and why diffusion is the mechanism that makes the thesis more severe rather than the counterargument to it. This series does its economic interior. The two read together.

The nine chapters in brief:

I. Energy and Climate — the climate objection to AI, taken seriously, and why the hyperscaler buildout aligns the political coalition for clean firm power rather than splitting it.
II. The Coasean Singularity — what happens to the inside of firms when transaction costs approach zero. Hollowing, not dissolution.
III. Baumol’s Revenge — the central chapter. The mechanism of the window, the Danish fiscal arithmetic, the manufacturing-capacity gate that bounds it.
IV. The Invisible Boom — why measured productivity lags. The J-curve, the consumption-channel mute, the Brynjolfsson–Acemoglu fork, and a side picked.
V. Winners Take Most — where the rents accumulate. Non-rival capital, the substrate stack, the Cochrane Euler decomposition, and the rate-regime question implicit in it.
VI. The Labor Question — displacement, reinstatement, a stance on flexicurity.
VII. Who Owns the Machine — capital-rent redistribution, sovereign-wealth-fund operationalisation, personal price discrimination, the libertarian objection and where it actually binds.
VIII. Creating Value — what the individual reader does. Age-stratified, interest-stratified, not reducible to the policy question.
IX. Endgame — the Romer loop, the Danish policy package, the open questions on which the thesis can be falsified again.

The window opens before the productivity that closes it arrives in the statistics. The capital that captures the rent in between owes nothing to the wage base under the income tax.

Leopold Aschenbrenner, Situational Awareness: The Decade Ahead, June 2024. The series treats Aschenbrenner’s capability-trajectory, intelligence-explosion, compute-and-energy, and US–China-security claims as the premise. What he leaves untouched — labour share, wage dynamics, fiscal pressure on welfare states, small-open-economy positioning, consumer-side rent capture — is the ground this series occupies. ↩

Footnotes