The Diamond in the Riverbed
All things are an exchange for fire, and fire for all things, as goods are for gold and gold for goods.
— Heraclitus, DK 22B90
Imagine a hunter close enough to a tiring gazelle that one more stretch of pursuit may bring it down. At the edge of a dry riverbed, light catches in a stone. If he stops, the meal escapes. He stops.
The stone is a diamond: carbon fixed under pressure into a lattice that will outlast him. It offers no food and little use as a tool. Carrying it back costs energy at the moment he has already spent most of what the hunt required. Yet he keeps it, accepting an immediate loss for the possibility that other people will recognize something in the object that the gazelle could never supply.
The choice does not prove that the stone will be valuable. Suppose the others glance at it and turn back to the fire. The expenditure remains real: the distance run, the meal surrendered, the burden carried home. The value does not appear. Work can produce a durable thing without producing a claim anyone else will honor, just as labor can produce a machine no one needs or a model no one will use. The mud pie objection begins here, before money or markets enter the story.
Nor would later exchange allow the hunter to name his own price. Another person might want the crystal because it catches light, because it is scarce, because it can be carried without decay, or because someone farther away has already agreed to trade for stones like it. Those judgments belong to a social world the object cannot create for itself. Expenditure matters because the desired arrangement is costly to find, make, preserve, or reproduce. It does not compel desire.
The diamond nevertheless differs from a promise made of air. Its hardness, refraction, and lattice can be tested. The tests establish properties of the stone and may distinguish it from other stones; they do not establish who found it, how it reached the exchange, or what anyone ought to give for it. Evidence travels farther than the person who spent the effort, but it carries only what the evidence can show.
By taking the stone back, the hunter places a present sacrifice before a future audience. He is wagering that the object will survive, that its properties will remain legible, and that other people will continue to care. Capital begins in that separation between the time of expenditure and the time of recognition. The larger the separation, the more the claim depends on institutions capable of preserving ownership, testing properties, and carrying judgments from one exchange to the next.
Nothing in the crystal sets a price. Its future value depends on a social judgment: that it is rare, durable, and desirable, and that this judgment will persist long enough for the object to be exchanged. The hunter's choice is therefore a wager on recognition. Work is spent now for a claim that can be redeemed only through other people.
The imagined choice isolates the problem this act will follow. Expenditure may produce an object with durable properties, but value appears only when institutions and people recognize those properties and decide what they warrant. The stone can outlast both the choice and the chooser. Later capital forms will repeat that temporal asymmetry in less visible materials.
The Thermodynamic Succession
An economy carries the results of work forward in time. Physics sets limits on that carrying, but economics begins where people decide which configurations are useful, who may claim them, and what they will exchange for them. Energy is necessary to production; it is never a sufficient account of value.
The earliest forms arrested decay directly. A clay-sealed granary in the Fertile Crescent stores solar energy as carbohydrate bonds, but decomposition begins almost immediately. A sealed granary slows the process long enough for the year's harvest to bridge the gap to next year's planting, and that bridge (the capacity to carry work forward across a season) is the foundation on which settled civilization was built.
Once surplus can be saved, it can be counted, and once counted it can be claimed. Seals, tallies, scribes, guards: this institutional machinery, emerging independently in Sumer, Egypt, the Indus Valley, and China, appears because stored work that lacks institutional protection leaks away. Energy that is not guarded dissipates, through theft, spoilage, vermin, flood, or the quiet diversion of surplus by those who control access. Institutions are the immune system of stored energy.
Metal solved constraints that grain could not address. A stamped coin is portable where grain is immobile, durable where grain rots, divisible where grain resists partition. When the kings of Lydia stamped electrum coins in the seventh century BC, they created an object that carried the claim of stored work across distances that moving physical commodities would have rendered uneconomic. A merchant in Sardis could hand a stamped coin to a merchant from Miletus, and the coin's metallic content, verifiable by weight and assay, substituted for personal knowledge. The coin did not eliminate trust. It reduced verification cost to the cost of weighing a small piece of metal, and the reduction was enough to enable coordination at scales that barter and grain-denominated credit could not reach. But coin traveled with the transaction. Metal is heavy, and the weight that guarantees its value also limits its range.
Credit broke the spatial bind. In medieval Venice, a written contract could pair a capital provider with a ship captain, money in one location funding risk and labor in another, because the courts of the Rialto made breach costly enough to substitute for physical presence. Promised future performance could fund present action; the enforceability of the promise was the mechanism by which capital could cross the Mediterranean without the merchant himself having to move.
Silos coordinate villages. Coins coordinate regions. Credit instruments coordinate empires. When the binding constraint shifts from preservation to transport to enforcement, the capital form that served the old regime becomes inadequate, and new structures crystallize around the fresh bottleneck.
England in the seventeenth century had built a commercial system of considerable sophistication: joint-stock companies, insurance markets, negotiable instruments, a banking infrastructure connecting London to the major trading centers of Europe. But the physical foundation remained organic. Wood for fuel, fodder for draft animals, crops for human labor: all drew on annual solar energy captured through photosynthesis, and the island's population was pressing against that ceiling. E.A. Wrigley named this the organic energy regime. The economy's useful annual work was bounded by the solar energy the island's acreage could capture, and no institutional cleverness could raise the ceiling without accessing a different energy source.
Coal was not simply better wood. It accessed a different energy stock entirely: not this year's sunlight but the Carboniferous period's, compressed by geological processes over three hundred million years into seams of concentrated chemical energy that could be burned faster than any living system could regenerate. A single coal-fired steam engine could do the work of dozens of horses, and it did not need to be fed, watered, or rested. The capital investments that the new energy throughput made possible (the railroads, the mills, the foundries) required institutional forms that the guild system and the cottage industry could not provide. The limited-liability corporation, the factory system, the wage contract, the industrial union: each emerged as a solution to coordination problems that the new energy regime created.
Institutional forms changed. The physics beneath them did not. Wealth is ultimately the capacity to do work against entropy: to organize matter and energy into configurations that serve human purposes and to maintain those configurations against the universal tendency toward disorder. Financial systems create claims on this true wealth, denominated in currencies and recorded in ledgers and traded on exchanges, but the claims themselves are not wealth. They are promises, and promises can multiply faster than the physical flows required to honor them. When the divergence grows large enough, correction arrives through default or inflation or restructuring, all of which amount to writing claims down to match what the economy can physically deliver.
Factor Prime
A hyperscale data center draws power at rates that would have supplied a small city in the industrial era. One hundred megawatts is common for a single facility; two hundred is no longer unusual; facilities drawing five hundred megawatts are under construction. Electricity flows in through high-voltage transmission lines whose capacity was originally built for aluminum smelters and steel mills. Heat flows out through cooling towers or liquid-cooling systems that pump chilled water through channels machined into the server racks. Between the inflow and the outflow is computation: the organized transformation of electrical energy into structured information.
Inside those racks, transistors switch billions of times per second, and each state transition dissipates a small quantity of energy into the silicon lattice as waste heat. The thermodynamic ledger is exact and unforgiving. Every inference, every training batch, every parameter update is purchased with electricity and paid for in heat. There is no computational free lunch. Energy that enters the facility as electrons in a wire leaves as heat in a cooling stream, and the difference between the two (the work that the electricity did before it became heat) is the computation.
Training a neural network is a search through a parameter space of extraordinary dimensionality, measuring the model's predictions against an objective at each step and adjusting weights to reduce the error. Gradient descent provides direction, but the space is vast and complex. What crystallizes at the end of training, if training succeeds, is selected information: a configuration of parameters that performs a specified task more reliably than a random arrangement would predict, a structure whose particular arrangement is improbable and whose improbability is the source of its value.
Once trained, the model can be copied at near-zero marginal cost and deployed millions of times simultaneously. But creation was expensive: computational work to search the parameter space, electrical work to power the search, physical infrastructure to generate, distribute, and dissipate the energy at scale. The model is energy and search, winnowed by selection, stored in weights. Like the diamond, its value derives from the improbability of its specific arrangement and the recognition that this particular arrangement, selected from the space of all possible arrangements by a process that cost real energy and real time, can do something that no random arrangement could do.
A diamond's observable properties can be tested: hardness, refraction, and lattice structure. Those tests establish properties of the stone. They do not establish its provenance, the labor conditions under which it reached the market, or the price anyone should pay; a synthetic diamond can share the relevant structure. A model can likewise be tested on a defined task, but the result establishes performance only on that evaluation and distribution. It does not prove who trained the model, how much work training required, whether the model was copied or distilled, or what the model is worth.
The stone did not prove its price. It proved the work.
But the analogy breaks where the economics diverge. A diamond cannot be copied. Its scarcity is geological and permanent. A trained model can be copied perfectly, at negligible cost, and the copies are functionally identical to the original. Scarcity resides in the process (the training run that cost millions of dollars and consumed megawatts of power) rather than in the artifact. Once the run is complete, the result can be distributed to anyone, and the distributor bears no cost beyond bandwidth. Fixed cost of creation remains high and is not recoverable if the model fails to find a market.
Factor Prime is this book's name for a proposed economic lens: energy structured through computation and disciplined by selection. Energy matters because computation is physically bounded. Computation matters because the expenditure changes decisions rather than merely producing heat or motion. Selection matters because expenditure alone creates no economic surplus; most searches fail, and some technically successful systems produce nothing anyone needs.
Capital and labor, the two factors around which industrial economics built its entire apparatus of theory, measurement, and policy, are becoming routing decisions over a shared substrate. A dollar invested in training a model that automates a task converts energy into what classical economics would call capital (a durable productive asset). The same dollar invested in running that model to augment a human worker's output converts energy into labor augmentation. But the physical substrate is the same: electricity, silicon, algorithms. The distinction between capital and labor, which organized everything from tax policy to labor law to class politics for two centuries, becomes unstable.
The worker who loses her job to automation experiences something categorically different from the investor who profits from it. When a firm can reallocate compute from training to inference in minutes, shifting resources from capital formation to labor augmentation and back again, the factor distinction that Ricardians and Marxists argued about for a century becomes a slider on a dashboard.
Yet the physics is indifferent to the politics. A system that consumes megawatts and produces nothing anyone will pay for has participated in Factor Prime's physics without participating in its economics. The energy and computation were real. The structure that resulted may have been thermodynamically deep, rich in the computational complexity Seth Lloyd defined as the minimum steps to produce the structure from randomness. But if no one will exchange anything for the output, the depth is a sunk cost and the structure is a computational mud pie.
A power plant's entropy production is coupled to a load that does useful work. A bonfire's entropy production dissipates into the night. A training run that produces a model people will pay to use is a power plant. A training run that produces a model no one wants is a bonfire. Thermodynamic signature identical. Economic outcome opposite.
What the Diamond Teaches
The diamond's lattice records a history of formation. Hardness, refraction, and adamantine luster let a trained examiner distinguish properties without taking the seller's word for them. Modern laboratories can establish much more, including whether a stone is natural or synthetic, but the inference always depends on a test, a comparison class, and an account of possible production routes. The lattice is evidence, not an oracle.
The stone can outlast the dynasty that mined it and the merchant who carried it. Its market value may not. Price depends on scarcity, fashion, control of supply, provenance, and the institutions through which buyers and sellers meet.
Monetary designs often seek a related property: evidence that cannot be altered cheaply by an issuer. Gold offers geological scarcity and chemical stability, yet it still requires assay, custody, and institutions of exchange. A diamond is not a purer currency; it is the clearer teaching object. Its properties can be inspected while its price, provenance, and justice remain unsettled.
A trained model admits the same limited form of inspection. Run it against a disclosed test set, measure the agreed quantities, and compare the result with the performance claim. The evaluation can refute that claim within its scope. It cannot prove training provenance or market value, and it cannot rule out a shortcut, copied weights, contaminated data, or a shift in the world beyond the test set.
But unforgeable structure does not settle the question of distribution. A diamond cannot be faked, but it can be hoarded. For the trained model, evaluation is available to anyone who performs it, but the model itself is available only to those who can access it, and access is controlled by whoever bears the cost of creation. Value therefore raises two questions: how is it produced? and who captures the surplus? The second is ultimately a question of governance.