An economy fully automated by AI and robots could keep growing after household demand disappears, according to a preprint that models a hypothetical economy after artificial general intelligence, or AGI. The model gives corporations ownership of conventional capital, machine agents and capacity to capture energy, while machines handle production, research, management and the fabrication of more machines. Within that setup, it reports a positive maximum growth rate even when human consumption is zero.
The result is conditional on a strong premise: the production function assumes no essential human input and sets the human labor share to zero. The paper is identified as a working paper that is not peer reviewed, and its illustrative growth path is explicitly not a forecast.
A demand loop with no households
The paper contrasts two ways of producing an economic agent. Human reproduction is represented with an approximately 20-year lag between investment and delivery of a productive unit. Machine agents can instead be fabricated; their modeled growth rises as unit costs fall along a learning curve and can be parallelized.
Within those rules, demand can close inside the corporate network. The paper presents a zero-human-consumption economy as a non-degenerate expanding system with positive, maximal growth, leaving household purchases outside the modeled demand loop.
The conclusion depends on full automation. If an essential task cannot be automated, the paper says growth would be pulled back toward the human-constrained rate.
The growth rate is a model result
In its illustrative calibration, the model uses a capital-output ratio of 3 falling toward 1.5, a share of output reinvested between 0.6 and 0.95, and depreciation of 0.1 per year. Before any contribution from automated research, the modeled growth rate ranges from 30% to 100% a year.
A theorem states that, under its stated conditions, productivity and modeled growth tend to infinity, leaving no path with a stable growth rate. The paper also imposes a physical ceiling: long-run growth is bounded by the growth of energy capture, so the runaway acceleration is treated as transitional.
Ownership becomes the welfare question
The paper's central welfare variable is the human ownership share. It is the financial link between humans and the corporate network: if the payout tied to that share shrinks faster than output grows, human consumption tends to zero while output tends to infinity.
The result is a sharp separation between GDP and welfare in the model. How machine agents are classified—property or persons—changes GDP's level and composition, but leaves modeled real dynamics, real-quantity growth rates and relative prices unchanged.
At the model's maximum growth rate, the interest rate equals the growth rate. Under positive human consumption, the human ownership share decays unless the economy operates inside its maximum expansion frontier or statutory transfers intervene.
The paper describes three possible terminal regimes—rentier, fully decoupled and socialized ownership—but does not choose among them.
A conditional warning, not a forecast
Those regimes are alternatives in the model, not selected outcomes. The machine-led growth result also rests on full automation, including no essential human input; if an essential task cannot be automated, the paper says growth is pulled back toward the human-constrained rate.
The energy ceiling is likewise part of the model, and the 30% to 100% annual path is presented as an illustration rather than a forecast.
Taken on its own terms, the preprint is a conditional warning about what GDP may miss: output can expand while human consumption falls when ownership claims shrink. It is a counterfactual study of post-AGI, full automation and alternative ownership regimes, not evidence that those conditions will be met or that any one regime will follow.
Paper data and sources
Original title: Growth Without Us: Machine Consumers, Corporate Circularity, and the Decoupling of GDP from Humanity after AGI
Authors: Sahil Sharma
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text