Two Computational Orders

What made war inevitable was the growth of Athenian power and the fear which this caused in Sparta.

An account of computation built only from prices misses the largest buyer, regulator, subsidist, and denier of access. States do not merely slow markets at their borders. They build grids and universities, finance fabrication, classify technologies, restrict exports, purchase systems, and reserve capacity for purposes no private return can price.

Computation has entered that field because its industrial base is concentrated and its uses bear directly upon military power, surveillance, scientific capacity, and economic administration. Chris Miller's history of the semiconductor industry shows how fabrication became a distributed achievement organized around extraordinarily narrow chokepoints.¹ No country commands the whole stack. Design tools, lithography, materials, fabrication, packaging, energy, and skilled labor are distributed across alliances whose commercial integration can be interrupted by political decision.

This changes the argument of the book. Physical throughput may be routable into cognitive work, but the route is neither naturally open nor governed by one price. A state can decide that a chip may not travel, that a model may not be trained, that a data center must serve defense before commerce, or that a capability must remain inside an administrative boundary. Scarcity then follows sovereignty as well as geology and engineering.

Two computational orders become possible. In one, access is allocated predominantly through markets. Firms assemble capital, purchase inputs, and deploy systems where expected returns exceed cost. In the other, strategic authority determines which actors receive chips, power, data, and permission. Neither order exists in pure form. Markets depend upon public infrastructure and law; states buy from firms and observe prices. The distinction concerns which institution prevails when commercial return and strategic purpose conflict.

The market order is not a natural default. Property, contracts, interconnection rights, export permissions, corporate forms, and currency are already public arrangements. Nor is administrative allocation simply the suspension of economics. Officials confront opportunity costs even when they refuse to express them in market prices. What changes is the objective: capacity may be directed toward resilience, military advantage, autonomy, prestige, employment, or political control rather than its highest private bid.

V/C does not determine this choice. It can still illuminate how easily a proposed use is evaluated once authority has selected it, but a planner may fund a low-V/C project for strategic reasons or forbid a high-V/C deployment for political ones. Likewise, the Joule Standard has only local force. An operator comparing flexible loads may observe a mining outside option; a state reserving power for a national program may prohibit that comparison from governing at all.

Chokepoints Become Borders

A concentrated production network can be efficient in ordinary time and coercive in crisis. Specialization lowers cost because each participant masters one difficult layer. The same specialization allows control of a layer to become leverage over the whole. A fabrication tool, process recipe, trained workforce, or power corridor may have no prompt replacement.

This is more than a supply-chain inconvenience. Denial can change who is able to experiment and therefore who learns. A jurisdiction cut off from frontier hardware may adapt algorithms to mature equipment, expand domestic production, steal designs, or redirect capital toward substitutes. Its systems will not simply be delayed copies. Constraints can produce different architectures, standards, and institutional relations.

The likely result is not one global computational economy separated by a temporary political wall. It is a family of partially incompatible orders. Models may be trained under different disclosure rules, deployed through different identity systems, connected to different payment rails, and made answerable to different principals. Technical standards will carry political assumptions about whose authorization counts and which records may cross a boundary.

Settlement will diverge as well. Permissionless networks, commercial stablecoins, bank money, and state digital systems make different trades among openness, control, recourse, privacy, and finality. Bitcoin supplies a vivid specimen of resource-secured permissionless settlement, but it does not follow that one geopolitical bloc will adopt it as official infrastructure. An open protocol can be used across rival jurisdictions, restricted within them, or surrounded by custodians that restore many of the intermediaries its base layer omits.

Proof-of-work matters here because it makes the cost of extending and revising a ledger state partly external to any one state's promise. That is a constitutional achievement, not immunity from politics. Mining can be regulated; network access can be filtered; custody can be concentrated; property claims can still depend upon courts. Confirmations increase resistance to revision without producing absolute finality or a remedy for every wrongful transaction.

Three Pressures

The first pressure is bifurcation. Rival states seek domestic capacity and reduce dependence upon suppliers they cannot control. Redundancy raises cost but may buy strategic resilience. As standards and supply chains separate, firms lose some scale while governments gain options.

The second is confrontation. A blockade, attack, sanction, or coercive interruption at a fabrication chokepoint could reverse the expected fall in computational cost. Capacity would be rationed. Training and inference would compete with intelligence, weapons, logistics, and critical administration. The book's physical account would become more visible at the precise moment its market sequence became less predictive.

The third is domestic control. Governments may treat large-scale computation as a utility, licensed activity, or sovereign resource even without international conflict. They might require reporting, reserve emergency access, subsidize favored uses, or condition deployment on conformity with public objectives. Such measures may correct private underinvestment in resilience, or place an infrastructure of thought under administrative command.

These pressures can coexist. A state may subsidize private fabrication, restrict foreign sales, and preserve competitive allocation at home. Another may tolerate open models while licensing the data centers able to train them. The economically important boundary will often lie inside the formal categories of public and private: who can direct capacity in an emergency, whose objective controls the queue, and who bears the cost of idle resilience.

The difference is observable. When prices rise, does capacity move toward the highest bidder or remain reserved for a designated purpose? Can a firm sell a capable model abroad? May operators choose among inference, training, and settlement work, or are some uses prohibited? Are deployment rights granted by general rule or administrative discretion? These decisions reveal the order beneath nominal ownership.

Rival Constitutions of Computation

The two orders also create different risks of capture. Market allocation can concentrate power in firms that control fabrication, cloud access, distribution, or data. State allocation can concentrate it in offices whose security claims resist review. A hybrid can combine both pathologies: public authority shields a private incumbent while private secrecy prevents public scrutiny.

Competition between orders may discipline them. It may also make each less corrigible. Governments can invoke foreign rivalry to subsidize domestic concentration, weaken disclosure, and treat dissent as delay. Firms can present their commercial position as national capacity. Strategic urgency then becomes the solvent of ordinary accountability.

The answer is not to pretend computation is nonstrategic. It is to govern strategic capacity without allowing emergency to become its permanent constitution. That requires explicit allocation rules, reviewable restrictions, time limits, institutional independence, and public knowledge of who can redirect the infrastructure. Resilience purchased through secrecy can leave a society unable to determine what it has protected.

Factor Prime survives this complication only in revised form. Energy, plant, and computation still constrain what can be done. Selection still distinguishes useful capability from dissipated work. But no physical account can derive the political purpose toward which capacity will be routed. Markets and states are rival mechanisms for making that choice, and each arrives already organized by law and power.

The industrial metabolism of cognition will therefore have more than one political anatomy. Some systems will be built to price opportunity. Others will be built to command it. The decisive struggle is not whether computation obeys physics—it does—but who is permitted to decide what its physical capacity is for.