Core Thesis
Durable economic gains from machine intelligence will not primarily accrue to the AI labs and model creators (“Genesis”). Models commoditize quickly, training costs explode, and frontier capability becomes a sovereign capability limited mainly to the US and China. Lasting value instead flows to scarce, controllable positions of economic power—the Commanding Heights—through which increasingly capable and widely diffused intelligence must pass to reach the real economy. These positions durably capture a share of the activity they enable; the more intelligence spreads, the more entrenched they become.
Structure and Main Arguments
Ravi (narrative): An Indian IT professional in Bangalore sees AI agents automate back-office work for American clients. Company margins rise while jobs vanish, contributing to rising youth unemployment and social unrest. Illustrates uneven diffusion and a widening K-shape.
Overture: Historical pattern repeated across revolutions (printing press, railroads, transistors, oil). Inventors rarely capture the bulk of the wealth; it settles with those who control the means of conduction and distribution.
Genesis / Intelligence Adolescent: The AI industry will keep spending on itself even as aggregate returns fall. Most current revenue still comes from early adopters plugging models into existing workflows. True broad diffusion is only beginning.
The Line: The US economy, household wealth, and credit are increasingly tied to the simultaneous rise of AI capex, stock prices, and model capability. Sovereign backstops are expected to sustain the trajectory.
Principalities and Powers: Only America and China can sustain continuous frontier model development. Intelligence splits into a commercial/diffused layer and a sovereign layer kept for statecraft.
Superscription: America’s greatest power is the Dollar, amplified by global stablecoin demand (“Neo-Dollar”). This creates new demand for Treasuries while owners of the economic channels collect inflation-sensitive tolls.
The World to Come: Diffusion begins in rich-world back offices (often located in developing countries). Citizenship, licensing, and guilds protect some workers; many others lose both jobs and traditional emigration paths. The K-shape widens further.
The Firm: Future firms consist of a small number of operators (compensated more like capital) directing work performed largely by models.
Alpha and Omega: Investors risk false security by clinging to last era’s Heights or simply financing this era’s inventions. The new Heights are still largely unpriced.
Closing: Technology has a directional force; the present is the end of one world and the beginning of another.
Characteristics of Commanding Heights
Scarce and controllable.
Sit on the vital path intelligence must travel to reach its utility.
Durably capture a share of the economic activity they enable (unlike temporary bottlenecks).
Function as tollbooths whose value multiplies as the underlying activity grows.
Not the invention itself, but the means of conduction/distribution.
Self-reinforcing: more activity strengthens the position and makes it harder to dislodge.
Investment Implications
The article argues that investors will tend to get the allocation wrong by either holding last era’s Commanding Heights or by financing this era’s inventions (the labs). The durable gains are expected to settle at the new Heights—scarce, controllable positions that meter the diffusion of intelligence—and these positions are described as still largely undiscovered and unpriced.
Implied orientation:
Be cautious of pure Genesis exposure (labs and pure model-training plays), as models commoditize and capital destruction is possible in that layer.
Be cautious of last-era Heights whose existing tolls intelligence can route around.
Actively seek emerging control points that meet the Height criteria—especially distribution and interface rails, payment and settlement infrastructure (particularly dollar-linked or stablecoin channels), access/metering layers, and proprietary activity-data inventories.
Treat identification of these positions as an ongoing active task rather than a settled conclusion.
Highest-potential categories among today’s companies (under the article’s filters):
Distribution/interface control: Apple, Microsoft, Meta (Google/Alphabet more qualified, given Search vulnerability).
Payment and dollar rails: major card networks, bank settlement infrastructure, or dominant stablecoin platforms.
Cloud/access platforms in their role as default integration and serving layers rather than pure training compute.
Platforms that accumulate hard-to-replicate activity data as intelligence spreads.
Pure AI labs and pure last-era toll businesses are lower-probability under the framework. Potential is not certainty; a position can lose Height status if intelligence finds a new route around it.
In short, the essay argues for reconstructed portfolios that seek the still-unpriced, scarce control points of intelligence diffusion rather than resting on either the inventors or the dominant positions of the prior era.
