An economy of corporate-owned AGI can grow explosively even if humans neither work nor consume—GDP simply becomes inter-corporate investment and machine operating demand; human well-being, however, reduces to a single variable: the fraction of corporate ownership humans retain, making ownership policy decisive for human survival in prosperity.

Main Finding

Sharma (Aug 2026, working paper) shows that a post-AGI economy in which corporations own reproducible AI/robot agents is internally coherent, can grow far faster than today, and need not collapse for lack of “human demand.” The economy becomes an inter-corporate, von Neumann–style expanding system in which (i) firms both produce and “consume” machine operating inputs, so demand closes without households; (ii) the demographic constraint on growth is removed once agents are manufactured rather than reared, allowing growth rates orders of magnitude larger and possible hyperbolic acceleration; and (iii) human welfare decouples from aggregate output except through a single state variable—the human ownership share εt—so the political/legal allocation of ownership determines whether humans benefit.

Key Points

• Consumer role is an accounting category, not a biological one:

  • Machine agents require flows (energy, compute, maintenance, upgrades) that function as demand. Whether these flows are booked as intermediate consumption (machines as property) or final consumption (machines as persons) changes GDP accounting but not the real physical dynamics (Lemma 2).
  • A closed inter-corporate economy that reinvests net output is the classical von Neumann expanding economy; with full reinvestment it attains maximal growth (Proposition 2).

• Removing demographic reproduction as the bottleneck:

  • Human reproduction technology imposes a low ceiling on agent-population growth (∼1–3%/yr); manufactured agents obey ordinary capital accumulation and can be scaled rapidly, limited mainly by fabrication throughput, energy capture, and learning-by-doing costs (Proposition 1).
  • Agent unit cost declines with cumulative production (Wright-law), so feasible growth rates rise over time; recursive improvements (agents improving agent production) can generate hyperbolic (super-exponential) phases.

• AK-style core and idea production:

  • With fully reproducible inputs (capital K, agents M, energy E) and substitution, the economy reduces exactly to an AK-like engine: Y = ˜A(t) X and growth depends on reinvestment share and the productivity of the reproducible core (Lemma 1).
  • Idea production (R&D) can be done by machine researchers; technology growth inherits the fabrication-limited rate, replacing demographic constraints of semi-endogenous models.

• Decoupling of GDP from human welfare via ownership:

  • All human stake in the autonomous corporate network is summarized by the ownership share εt. In the maximal-growth frontier equilibrium the interest rate equals growth (r = g); if humans consume a positive fraction of their wealth, εt decays exponentially at that same rate (golden-rule decoupling theorem).
  • Three terminal regimes are possible and are policy-determined, not technology-determined:
  • Rentier post-scarcity (ε bounded away from 0): tiny ownership shares deliver large human consumption because output grows hyper-exponentially.
  • Full circular decoupling (ε → 0): corporate retained earnings, buybacks, and inter-corporate holdings extinguish human claims while output diverges.
  • Socialized ownership: states or public funds hold ε on citizens’ behalf.
  • Normative conclusion (narrow): after AGI, employment policy is largely irrelevant; ownership policy is the central instrument for human welfare.

• Physical ceilings exist:

  • Speed and number of doublings are ultimately bounded by energy capture, materials, and thermodynamic/computational limits (Landauer-type bounds), but these constrain how far growth can go, not the basic mechanism that permits much faster growth.

Data & Methods

• Nature of the paper: theoretical, continuous-time model; working paper (not peer-reviewed); no empirical dataset.
• Core formal tools and assumptions:
  • Full automation assumption: machine agents are perfect substitutes for human labor in all tasks.
  • Production function: Yt = At Kt^α Mt^β Et^{1−α−β} (α,β>0, α+β<1).
  • Capital accumulation:
    • Conventional K accumulation: standard investment-depreciation law.
    • Machine agents M: ˙M = IM/cM(t) − δM M with cM(t) = c0 Q_t^{-γ} (Wright-law learning curve).
    • Energy capacity E: accumulable with its own unit cost path.
  • Idea-production (R&D): ˙A/A = χ M_R^λ A^{ϕ−1} (machines as researchers; Jones-style idea production).
  • Reduction to AK form: define composite reproducible stock X ≡ K + cM M + cE E; yields exact linear relationship Y = ˜A(θ) X and standard AK growth dynamics.
  • Demand modeling: machine operating bundle xM = (energy, compute, bandwidth, maintenance, upgrades); owners set budgets and agents’ operating demand solves a constrained maximization (machina economicus), giving standard demand properties.
  • Identification with von Neumann expanding economy: the inter-corporate trading of operating flows and capacity purchases creates a self-expanding activities matrix and a balanced-growth maximal solution.
• Analytical approach: propositions, lemmas, and comparative statics; characterization of steady/frontier growth, dynamics of εt (human ownership share), and classification of terminal regimes. Discussion of how legal/accounting classifications affect measured GDP but not physical flows.

Implications for AI Economics

• Rethink demand-side objections to full automation:

  • The classic “who will buy the output?” underconsumption objection is formally rejected: firms can be the buyers. Policy debates must move beyond relying on household demand as an automatic insurance against human irrelevance.

• Ownership becomes the primary policy lever:

  • Because human welfare depends only on εt, post-AGI policy should focus on ownership allocation, governance, and institutional rules (dividend regimes, buyback/retained earnings rules, limits on inter-corporate cross-holding, public equity stakes, or outright socialization) rather than employment policy.
  • In practice, law and initial distributions (who owns the machine-producing firms at takeover/transition) determine whether humans obtain rents from autonomous growth.

• Measurement & accounts:

  • National accounting conventions will matter for observed GDP composition (intermediate vs final consumption of machine operation), but not for real resource allocation. Policymakers and researchers should expect GDP to be a poor proxy for human welfare in a post-AGI world; ownership-adjusted measures of human claims and consumption are essential.

• Growth and risk trade-offs:

  • Potentially enormous and fast growth (fabrication-limited, learning-curve amplified, recursive) increases stakes: faster growth raises the speed at which ownership shares can be diluted and accelerates political-economy transitions. This heightens urgency for pre-AGI design of institutions governing ownership and corporate behavior.
  • Although growth can be hyperbolic in theory, physical limits (energy, thermodynamics, materials) and institutional frictions will bound realizable trajectories. Calibration of these bounds is a priority for future work.

• Research and policy agenda suggested by the paper:

  • Calibrate machine reproduction costs, energy constraints, and feasible fabrication throughput to quantify plausible growth rates and transition speeds.
  • Model political-economy processes governing initial allocation of corporate equity, legal personhood of machines, dividend and buyback rules, and public-ownership options.
  • Re-evaluate safety and distribution policy trade-offs: policies that deliberately reduce firms’ maximal feasible growth (e.g., throughput caps, higher extraction/taxation on machine outputs) can preserve εt but may slow total growth—this becomes an explicit policy trade-off.
  • Redesign welfare metrics: track human ownership shares, convertible claims on autonomous production, and direct measures of human consumption rather than relying on GDP alone.

Limitations to bear in mind - Strong assumptions: perfect substitution by machines, firms maximizing net-worth growth, no explicit political resistance to expropriation or enforced distribution rules, and deterministic continuous-time dynamics. - No empirical calibration or stochastic modeling of transition dynamics; outcomes depend heavily on initial conditions and institutional choices, which are not endogenized here.

Bottom line Sharma formalizes a sharp scenario: an autonomous, fast-growing inter-corporate economy is internally consistent and can leave humans materially dependent only on the legal and political allocation of ownership. The paper reframes post-AGI policy from employment/consumption preservation to proactive ownership and institutional design.