Generated by Codex with GPT 5.6 Sol High

Techmeme surfaced Will Manidis and Joe Roberts’s October 29 essay, The Commanding Heights. It asks a sharper question than whether artificial intelligence will create enormous value: who will be able to collect a durable share of that value after capable models become cheap and widely available?

The authors’ answer is that the laboratories building frontier models may not be the long-term winners. They compare AI with earlier general-purpose technologies whose inventors changed the world without capturing most of the resulting wealth. The railroads made distant land, oil, manufacturing and trade more valuable; the strongest fortunes often belonged to those who controlled routes, finance or distribution rather than locomotive design. In the same way, the essay argues, AI’s largest returns may migrate from model production to the scarce channels through which intelligence reaches the rest of the economy.

From invention to tollbooth

The essay calls the model-and-compute industry “Genesis.” This includes frontier labs, hyperscalers, chips, data centers and the infrastructure that serves tokens. Genesis can keep advancing even if investment returns deteriorate. Model capability may rise, token prices may fall and society may receive more useful intelligence, while competing suppliers spend so aggressively that much of their own capital is destroyed.

That distinction is central. A technology can succeed as technology while disappointing its financiers. The authors expect open-weight Chinese models, rapid imitation and falling inference costs to compress the margins of companies selling intelligence directly. Any business whose advantage depends only on today’s model price or capability gap therefore looks temporary.

A true “Commanding Height” is different from a temporary bottleneck. It is a scarce, controllable position that takes a toll on expanding activity and becomes more entrenched as that activity grows. Search advertising was such a position in the internet era: countless businesses invested in websites and online commerce, while the dominant gateway collected a share of their demand. The essay looks for analogous gateways in an AI economy—distribution networks, payment rails, regulated access, identity, proprietary workflow channels and other positions that models must pass through to create economic value.

The line becomes a state interest

The authors describe AI capability, infrastructure spending and US asset prices as parts of one rising “line.” Major technology companies fund the buildout, their valuations support household and institutional wealth, and credit markets finance the next round of construction. As these exposures become more tightly coupled, keeping the system moving ceases to be only a private investment decision. It becomes a political objective.

From there, the essay predicts a split between three layers of intelligence. Cheap commodity models spread globally. A commercial frontier sells stronger capabilities at a premium. A sovereign frontier is developed or reserved for intelligence, military and state purposes. The United States and China, in this account, are the only countries able to sustain the capital, energy, chips and institutions required to advance that sovereign frontier continuously.

The two systems would exercise control differently. China can tolerate low-priced open models because its state does not need laboratory valuations to support household retirement accounts or public markets. The United States is more dependent on private firms and capital markets, but national-security needs would still give Washington increasing influence over laboratories and infrastructure. Neither country would need to announce formal nationalization. Funding, procurement, regulation and security obligations could produce effective state control while ownership remains nominally private.

The Neo-Dollar and the widening K

The essay’s most distinctive financial claim concerns what it calls the “Neo-Dollar”: dollar-denominated stablecoins used by people and software agents around the world. If AI makes digital commerce cheaper and more pervasive, the authors expect more transactions to flow through programmable dollar rails. Stablecoin issuers would hold dollar assets, including government debt, while payment and distribution companies would collect fees on activity that originated far outside the United States.

This creates a proposed feedback loop. AI expands digital economic activity; dollar rails settle that activity; demand for dollar assets helps finance the American state; and the owners of the rails gain a tollbooth whose traffic rises with AI diffusion. In the authors’ framework, this channel may be more durable than any one model laboratory’s technical lead.

The labor counterpart is a widening “K.” AI first automates work that is already digitized, measurable and delivered across borders. That makes outsourced back-office work in lower-income countries especially exposed. The essay’s fictional Indian manager, Ravi, supervises agents that gradually absorb his team’s work and then his own. Productivity gains appear as higher margins and US-listed equity value, while the lost job and social disruption remain in India.

Workers in richer countries may receive more protection through citizenship, professional licensing, regulation and political bargaining. Those defenses could preserve local jobs while removing an earlier development path for poorer countries: supplying increasingly skilled services to wealthier markets and using migration as an escape valve. The essay therefore expects AI abundance to coexist with sharper inequality between countries, within countries and within firms.

Firms after labor leaves the center

The authors also forecast a new corporate form. Industrial firms combined owners, managers and large employee organizations. The AI-era firm may instead consist of a small number of operators directing model labor, with those humans paid and governed more like capital partners than conventional employees. Venture capital becomes closer to trading in fast-changing technical advantages; private equity becomes a mechanism for inserting AI into existing businesses; and large firms increasingly perform functions once associated with the state.

This shift matters for investors because yesterday’s Commanding Heights can look permanent just before a new system displaces them. Railroads, newspapers and earlier information gateways once appeared to possess unassailable economics. The essay warns that simply owning today’s dominant technology companies, or financing frontier-model startups, may miss the next points of control. The real opportunity would be to identify the still-cheap channel that eventually taxes the entire AI economy.

A framework, not a forecast model

The essay is strongest as a way to separate invention from value capture. Falling model costs do not imply falling economic importance; they may instead push profits toward distribution, trust, regulation and settlement. Its distinction between a temporary shortage and a self-reinforcing tollbooth is also useful for evaluating AI businesses.

Its sweeping predictions deserve more caution than its confident style allows. Historical analogies show that value can migrate, but they do not identify where it must go next. Chinese open models may compress prices without eliminating laboratory power. Stablecoin growth does not automatically produce unlimited global demand for dollars or government debt; regulation, currency controls, competing payment systems and political resistance all matter. Labor displacement depends on adoption speed, new demand, institutional choices and the quality of work that agents can perform reliably. The essay often turns these contingencies into inevitabilities.

It also stops short of naming its most important investment examples, replacing two prospective Commanding Heights with redacted placeholders. That protects the authors’ intended portfolio insight but makes the central claim difficult to test. Readers receive a vivid theory of what a durable position should look like, not evidence that any particular company already possesses one.

The lasting question is nevertheless valuable: if intelligence becomes abundant, which scarce institutions will every useful deployment still need? The companies and states that control those institutions may capture more of AI’s economic surplus than the organizations that made the models possible.