An essay by Dean Gallagher
Draft · April 2026
I. The Constraint That Just Lifted II. The Bottlenecks III. The Trade I Already Made IV. The Pattern That Recurs V. The Capstone
Chapter I of V

Beyond Humanity.

For all of recorded human history, every economic model has shared a single, unspoken assumption — that the labor force grows at the rate of human births. That assumption is no longer true. The argument of this essay is that the assumption is being broken in real time by four converging mechanisms — and that the substrate of an economy whose participants are no longer constrained by the rate of human birth is being built on quarterly income statements right now.
Document
Beyond Humanity
Movement
I of V
Subject
The Bottleneck Decade
Window
2026 – 2032

Every economic model in history rests on the same hidden assumption. The labor force, the consumer base, the population of researchers, the count of decision-makers — all of it grows at the rate of human birth.

That ceiling has held since the agricultural revolution. It survived the steam engine. It survived electrification, the assembly line, the integrated circuit, the internet. Each of those technologies expanded what a single human worker could produce in an hour. None of them changed the number of workers.

Beginning in 2024, that assumption broke.

The break is not a forecast. It is not a thesis about what artificial general intelligence will do in some indefinite future. It is the empirical record of what is already happening, on operating income statements, in 13F filings, in every quarterly capex disclosure from every hyperscale cloud provider in the United States. The economy is acquiring participants who are not human.

The first wave is digital. 51% of large enterprises now have autonomous AI agents in production, generating an average 171% return on investment with a median payback period of 8.3 months. The second wave is physical. A humanoid robot, deployed at scale in a Western manufacturing facility in 2026, costs between $3 and $8 per hour to operate. A human worker costs between $22 and $40. The third wave is computational, and it is recursive: the agents are starting to do the research that builds the next generation of agents.

None of these participants requires a birth certificate. None of them ages out of the labor force. None of them files a tax return, votes, or counts toward gross domestic product the way an employed human counts. They consume electricity, semiconductors, networking bandwidth, and capital. They produce output. They are, in the dry vocabulary of economic measurement, actors.

For the first time, the size of the economy is no longer constrained by the size of humanity.

The labor force just stopped being a population-bounded variable.
Every forecast that assumes ~8 billion human consumers is structurally wrong from 2027 onward.
A trader saw it before economists did

The trade was already placed.

This document exists because Dean Gallagher recognized this break, in his personal capital account, beginning in January 2023. Over the subsequent forty months he compounded a personal brokerage account through positions concentrated in the four themes that this thesis identifies as the mechanisms by which the human-population ceiling lifts: artificial intelligence, semiconductors, robotics, and space.

Three years of his trading record — every transaction, every monthly statement, every drawdown — has been independently reconstructed and is reproduced in Chapter III. The record documents capital deployed, conviction held through a 36.5% drawdown in the spring of 2024, and a track record produced in the most rigorous environment in which one can compound capital: a personal account, with no fees, no investors to soothe, and no permission to be wrong.

This essay is the articulation of a strategy that has already been run with capital at risk. The thesis was tested in real time. The trade is live.

This document is the full articulation of why.

Figure 01 · Historical infrastructure mobilizations

AI infrastructure capex now exceeds the New Deal — and is projected to exceed the Gilded Age railroad buildout by 2030.

Annual peak spending in 2025 dollars, expressed as a percentage of contemporary GDP. The AI buildout has already eclipsed the dot-com telecom bubble in both absolute terms and as a share of national output, and is on a trajectory that compares only to wartime mobilization or 19th-century continental industrialization.

40% 30% 20% 10% 0% % OF GDP AT PEAK 37.8% 12.3% 7.7% 6.0% 2.8% 2.0% 1.6% 1.2% 0.9% 0.7% WWII 1944 WWI 1918 New Deal 1936 Railroads 1870 AI proj. 2030 Highways 1964 AI now 2024 Dot-com 2000 Manhattan 1945 Apollo 1966 By 2030, projected to exceed every peacetime mobilization since the railroad era.
Sources: Tunguz (LLM Impact on GDP), Bureau of Economic Analysis, Federal Reserve Bank of San Francisco, McKinsey ($7T to scale data centers, 2024), Goldman Sachs (Powering the AI Era), historical CRS reports. AI 2024 figure reflects total US AI infrastructure capex; AI 2030 projection assumes OpenAI's stated infrastructure plan at 30% market share.
The shape of the rotation

Four mechanisms. One break.

Population-bounded economic growth ends in four specific ways, and the four ways are the four themes that organize this fund. Each is a distinct mechanism by which the human-population ceiling on the economy lifts. Each can be measured, sized, and invested in directly through public equity markets today.

I

Artificial Intelligence

The agents themselves. The first new economic actors in the history of recorded labor.

II

Semiconductors

The substrate of every agent's cognition and every robot's motor control. The chokepoint through which every dollar of AI revenue flows.

III

Robotics

The embodiment of the agents. The mechanism by which the digital labor force captures the physical labor market — a market measured in tens of trillions of dollars annually.

IV

Space

The expansion vector. The infrastructure that multiplies addressable market beyond Earth's surface, and the substrate of the persistent intelligence layer above it.

None of these themes is novel. Each has been written about exhaustively in trade press, equity research, and venture pitch decks for at least three years. What has not been widely articulated is that they are the same thesis — four mechanisms by which the human-population ceiling lifts — and that they compound on each other rather than competing. Capital allocated against any single one captures a fraction of the rotation. Capital allocated against all four, with discipline, captures the full structural break.

Capital that misses all four spends the next decade benchmarking against an index that no longer reflects the underlying economy.

We are not investing in artificial intelligence.
We are investing in the substrate of an economy whose participants
are no longer constrained by the rate of human birth.
Why this decade and no other

The window is 2026 to 2032.

Structural breaks of this magnitude generate a narrow window during which capital allocated correctly produces returns disproportionate to the underlying risk. The window opens when the technology becomes economically viable but before consensus prices it in. The window closes when the rotation completes and the new constants — agents as economic actors, robotic labor at scale, $7 trillion in deployed compute infrastructure — become the boring baseline of every analyst's spreadsheet.

Three converging trendlines identify the window for the present rotation as approximately 2026 through 2032.

The capex acceleration is happening now.

The five largest U.S. cloud and AI infrastructure providers have committed to between $660 and $690 billion of capital expenditure in 2026 alone — nearly double 2025's level of $360 billion. Meta's 2026 capex by itself exceeds the entire annual capital spending of the U.S. semiconductor industry and rivals the spending of the entire utility sector. McKinsey models global AI-specific data center capital expenditure of $5.2 to $6.7 trillion through 2030, a sum that exceeds the gross domestic product of every country on Earth except the United States and China.

$660-690B in 2026 alone
Combined 2026 capital expenditure committed by the top five U.S. hyperscale cloud and AI infrastructure providers — roughly double the prior year's level. The acceleration is not a future event.

The capability curve has steepened.

Compute deployed to train frontier artificial intelligence models has increased by approximately 4.5× per year since 2010. Global AI compute capacity is doubling every 7 months. Frontier laboratories are explicitly targeting training runs of 2 × 10²⁹ floating-point operations by the end of the decade — the same magnitude leap from GPT-4 that GPT-4 represented over GPT-2. The frontier-laboratory roadmaps published by OpenAI, Anthropic, and Google DeepMind converge in their independent forecasts: the deployment of fully autonomous AI researchers, capable of conducting end-to-end machine-learning research, is targeted for the 2026-to-2028 window.

The arrival of automated AI research is significant not because it produces a science-fiction event, but because it changes the rate at which the underlying technology improves. Algorithmic progress, historically gated by the supply of human researchers (which grows at one to two percent per year), becomes gated by the supply of compute (which currently doubles every seven months). The function that connects capital input to capability output steepens.

The physical infrastructure is scarce.

Constraints on the rate of buildout are not financial. They are physical. ASML, the sole supplier of extreme ultraviolet lithography systems on Earth, will deliver a total of 66 EUV scanners globally through 2027 — a hard ceiling on the rate at which leading-edge semiconductor capacity can scale. Large power transformers required to step grid voltage to data center scale carry lead times of 128 weeks; generator step-up units of 144 weeks; some specialized units up to 48 months. The PJM Interconnection wholesale capacity market, which prices power for the eastern United States, has seen its clearing price rise 833% in two years and now sits at the federally imposed price ceiling.

These constraints generate the second-order investment opportunity. The companies that own the chokepoints — the lithography monopolist, the high-bandwidth memory oligopoly, the transformer manufacturers, the secured grid interconnection rights — capture the economic rent of the rotation regardless of which specific applications win at the user-facing layer.

What this document is, and is not

A thesis, not a forecast.

The forecasts that govern most investment decisions are forecasts of human behavior. Consumer demand. Labor productivity. Election outcomes. The diffusion of innovation through populations of human adopters. The professionals who produce these forecasts have refined their craft for decades, and on any reasonable time horizon, they are excellent at their job.

This document is not a forecast in that sense. It does not predict whether or when artificial general intelligence arrives, whether the Federal Reserve raises or cuts, which administration takes power in 2028, or which specific company captures the largest share of the cloud market in 2031. The economic models that produce those forecasts will, on most questions, outperform any framework offered here.

The argument in this document is structurally different. It is the claim that the underlying object being forecast — the size of the economy — has a new term in the equation, and that the forecasters have not yet incorporated it. The number of human consumers, workers, and decision-makers will continue to grow at approximately the rate of human birth. The number of non-human economic actors will, beginning in 2026, grow at the rate of capital deployment.

That second number, by 2032, will exceed the first. The total addressable market of every business in the world will be reset accordingly. The funds that own the substrate of that reset will compound. The funds that benchmark to the prior assumption will not.

The remaining six chapters of this document make the specific case for how a concentrated, four-theme, long-biased equity book should be constructed to capture this rotation. They detail the bottlenecks that produce the highest economic rent. They walk through Dean Gallagher's three years of personal trading as the proof of conviction. They engage honestly with the historical record of how thematic concentration funds have failed, and design the structure to survive what historically destroys books like this one. They state terms.

A reader who finishes the document and disagrees with its central claim is, by design, not the right limited partner for this fund. A reader who finishes it and recognizes the rotation as already underway will, we expect, want to know the specifics of how to participate.

What follows is the specifics.