\section{Introduction}
\label{sec:intro}

\citet{Restrepo2025} describes an economy in which artificial general intelligence has arrived: every type of
work that people do can also be done by computers, at a cost in compute that is finite and known, and the stock
of compute grows without bound. His conclusions are stark and, within his premises, exact. All work that is
essential for growth is eventually done by compute; output becomes linear in compute; wages converge to the
value of the compute that a worker saves; and the share of labor in income converges to zero while labor's
income does not fall. ``The economy keeps growing; we stay in place.'' The chapter is deliberately silent about
the other side of the national accounts. Compute is an endowment that nobody buys, depreciates or finances;
nothing is said about what the income that leaves labor is worth as an asset, who owns the claims to it, or
how that ownership evolves. This paper is about that other side.

We ask three questions. First, does the income that leaves labor go to the owners of compute, as the source
concludes, or to the owners of what compute cannot reproduce --- structures, land, energy sites? Second, what is
that income worth: does ``all income accrues to compute'' mean that the value of compute capital, or of wealth
in general, explodes? Third, how does the ownership of wealth evolve across households, and what do the policies
that the source mentions in passing --- universal dividends, treating compute ``as a public or semi-public
resource, akin to land'' --- do in general equilibrium?

To answer them we keep the source's task block intact and change one thing in production: compute is a
reproducible capital good with a price that falls over time, a depreciation rate, an energy requirement, and
complements --- structures, land and energy --- that it cannot replace and that enter output with an elasticity of
substitution that may be above or below one. Everything on the household side is new: a two-class economy whose
capitalists' Euler equation prices every claim, and a heterogeneous-agent economy in which households with
different skills, discount factors and productivity histories hold the claims through a fund. We solve the model
in three layers of increasing size: the exact replication of the source (Section~\ref{sec:benchmark}); a
limit economy with one complement in which eight propositions are proved (Section~\ref{sec:theory}); and a
quantitative economy with four capital stocks, energy and housing, calibrated to the United States in 2026 and
solved along perfect-foresight paths to 2300 under a catalogue of conditional scenarios
(Sections~\ref{sec:calibration}--\ref{sec:robustness}).

\paragraph{Findings.} The labor-share result of the source is more robust than its author claims, and its wage
result is less. The labor share goes to zero under any constant-returns production structure, with or without
complements, with compute as an endowment or as a priced capital good, because the only thing it needs is that
labor and compute be perfect substitutes within tasks (Proposition~\ref{prop:laborshare}). The level of wages, by
contrast, depends on compute remaining scarce. If the task composite and the physical complements are gross
complements ($\sigma<1$) and compute becomes cheap, the price of compute goes to zero and so do wages --- in level,
not only as a share --- and the income that labor loses goes to the owners of the scarcest complement
(Propositions~\ref{prop:regimes}--\ref{prop:moore}). With a floor on the compute price, wages converge to the
user cost of the compute that replaces the worker, $w_s=(r+\delta_Q)\bar p\,\CEUt(s)$: ``the rental price of
your replacement'' (Proposition~\ref{prop:bgp}). An energy bottleneck keeps the labor share positive
(Proposition~\ref{prop:energy}).

On valuation, the income share and the market value of a claim are different objects. On a balanced path
the wealth-to-income ratio is $\sum_ks_k/[(1+r)(1+\pi_k)-(1-\delta_k)]+s_T/(r-g)$: a claim that depreciates at
25\% a year and cheapens at 15\% a year --- compute --- is capitalized at a price--earnings ratio of about two,
against eight for structures and fifty for land (Proposition~\ref{prop:wealth}). Capital's share can rise while
wealth-to-income falls. And a growth boom raises the real rate through the marginal investor's Euler equation,
so that valuation ratios of long-duration claims \emph{fall} during the boom when the elasticity of intertemporal
substitution is below one (Proposition~\ref{prop:discount}).

Quantitatively, in the baseline scenario --- capability improving at 10\% a year, compute prices falling at 15\%
a year to one hundredth of today's level, energy efficiency gains with a floor --- the routine task is automated
by \AutoYearOne{} and the expert task by \AutoYearFour. The labor share falls from 0.56 to \LSTwoZeroFiveZero{}
in 2050 and below one in a thousand after 2075. The lower skill's wage falls in steps as the source's
compute-binding transition predicts; the higher skill's wage rises to \wHTwoZeroFiveZero{} times its 2026 level
while its expert task is the last one that compute cannot do cheaply, then collapses --- the source's
technology-binding transition. Output rises \GDPTwoZeroFiveZero-fold by 2050 and \GDPTwoOneZeroZero-fold by 2100,
after which growth returns to its trend because land binds. The real return on wealth rises to \rPeak\% in
\rPeakYear{} and the wealth-to-income ratio \emph{falls} from \WoverGDPbase{} to \WYmin{} in \WYminYear{} while the
capital share rises; it then climbs to \WYTwoOneZeroZero{} by 2100 as land, which cannot be produced, captures
\LandShareTwoOneZeroZero{} of income. The market value of compute never exceeds one fifth of GDP. The decomposition
of 2026 valuations into cash-flow and discount-rate news shows the two channels to be of the same order, with
opposite signs. In the heterogeneous economy the wealth Gini rises from \GiniBase{} to \GiniTwoOneZeroZero{} by 2100;
a citizen fund holding one fifth of all claims keeps workers' consumption share bounded away from zero; a compute
tax raises wages one-for-one with the tax-inclusive user cost at a cost in output; a land tax redistributes
without changing allocations; an AI monopoly protects wages the way a compute tax does, with the difference that
the rents go to the monopolist.

\paragraph{What this paper is not.} It is not a forecast. Every scenario input --- the rate of capability
progress, the compute price floor, the energy-efficiency floor, the terminal compute requirement of each task ---
is documented as such, and the level of wages after 2060 is proportional to the product of two of them. It is not
a model of the arrival of AGI under uncertainty: the paths are perfect-foresight equilibria, there are no risk
premia, and the stochastic-arrival exercise of Section~\ref{sec:robustness} is a labelled approximation. And it
does not model endogenous innovation, the creation of new human tasks beyond one reinstatement scenario, or
nonhomothetic preferences; Section~\ref{sec:conclusion} lists what those would change.

\paragraph{Organization.} Section~\ref{sec:literature} places the paper in the literature.
Section~\ref{sec:benchmark} reproduces and audits the source. Sections~\ref{sec:model}--\ref{sec:assetpricing}
present the quantitative model, its households and its asset prices. Section~\ref{sec:theory} states and
discusses the propositions, with proofs in Appendix~\ref{app:proofs}. Section~\ref{sec:calibration} calibrates;
Section~\ref{sec:results} presents the baseline; Section~\ref{sec:robustness} the scenarios, the heterogeneous
economy, policy and welfare; Section~\ref{sec:empirical} the empirical discipline and a falsifiable short-run
design; Section~\ref{sec:conclusion} concludes. The replication package reproduces every number with one command.