Trillions
of Markets

There was a time when “the market” meant a room. A pit in Chicago, a ring in London, a floor in New York where people in bright jackets shouted prices into the air as if civilization turned on whether wheat closed up three cents. Markets were loud, physical, countable things. You could point at one. You could regulate the room and tally the goods inside it: corn, oil, gold, the shares of the largest firms, the debt of the most creditworthy states.

Then markets became screens. Then screens became APIs. Then the APIs acquired agents. And somewhere in there the line between “the market” and “the world” got hard to draw, because the thing in the room had never been the market in the first place. The room was scaffolding. The market was the procedure the room performed: take a claim on some uncertain future, find someone who values it differently, agree on a price, settle. A pit is one way to run that procedure. It was the expensive interface we built when matching, trust, and settlement were hard.

Follow that reframing to its conclusion and the central fact about markets stops being architecture or regulation and becomes cost: how expensive it is to specify a claim, read the world, find a counterparty, quote a price, and move the money. For all of recorded history that cost was high, so the procedure was rationed to risks large enough to amortize it. That era is ending. The cost is falling on every margin at once, and when it does the procedure stops being a place you go and becomes something closer to a utility — ambient, mostly silent, running on everything.

A pit was not the market. It was the market’s expensive interface.

One inequality, and the resolution of reality

Start with the inequality. A market appears wherever the surplus a creator can capture from intermediating an exposure exceeds the cost of defining, verifying, pricing, and settling the claim that carries it. Hedging, financing, speculation, price discovery, allocation — these are the sources of surplus. Specification, data, legal form, collateral, market-making, settlement, fraud, disputes — these are the cost. For almost all of history the cost side won, so only the largest, most standardized risks ever cleared the bar: staple commodities, sovereign debt, the equity of great corporations. Everything else stayed trapped inside private ownership, opaque underwriting, or no transaction at all. The exposure was real. The market for it was priced out of existence by its own overhead.

That inequality is not the novelty; it is the frame. The thesis is what happens when many of its cost terms collapse together: market creation stops moving one product at a time, liquidity starts scaling through shared risk engines instead of isolated crowds, and value migrates away from venues toward the machinery that verifies reality, routes trusted agents, and decomposes risk.

This is an old idea running in fast-forward. Ronald Coase taught that the line between what happens inside a firm and what happens out in a market is set by a single quantity — the cost of using the price mechanism. Where transacting is expensive, activity huddles inside firms, long-term contracts, and unpriced relationships; where it is cheap, the boundary moves outward and the price system takes over more of economic life. For most of a century that boundary crept. Shahidi, Rusak, Manning, Fradkin, and Horton have called this a Coasean singularity in their work on AI-agent market design: the point at which using the price mechanism becomes cheap enough that the boundary runs away. This essay takes the same cost-collapse premise somewhere else — into the architecture of liquidity and the proliferation of tradable claims.

There is a cleaner way to see what the cost collapse really does. The number of distinct prices an economy maintains is the resolution at which it renders its own reality — how finely it has chosen to measure and value the world. Old finance ran at miserably low resolution: a few thousand continuously quoted instruments standing in for the entire risk surface of a planet of eight billion people, three hundred million firms, and a near-infinity of contingent futures. This is not a metaphor reaching for color. We have watched resolution increase before. When U.S. equity markets went from fractions to decimals in 2001, the minimum tick fell from an eighth of a dollar to a penny, spreads compressed, and prices got sharper overnight. That was a resolution increase on one axis of one market. The claim here is that the same increase is now coming on every axis of every market simultaneously: more instruments, finer claims, shorter horizons, more local contexts.

By a market I do not mean a room, a venue, or a deep pool of attention. I mean a claim that can be continuously repriced: a payoff rule with enough specification, verification, collateral, and liquidity to quote it against the world. We will not reach the theorist's dream of a price for every possible state of the world; most of that space is unmeasurable, illegal, redundant, or uneconomic. But we will approach a vast admissible slice of it far faster than anyone schooled on the old cost structure expects — and the count of live, repricing markets at the end of that process is not in the thousands or the millions. It is plausibly in the trillions.

The number of prices an economy keeps is the resolution at which it renders reality. The resolution is about to increase by orders of magnitude.

Three costs fell below the line

The obvious objection deserves an answer before anything else, because it is the strongest one. You have described, the skeptic says, a forty-year decline in transaction costs — electronic trading, decimalization, the internet, the ETF — and you have dressed an extrapolation up as a revolution. This is a trend, not a phase change.

It is a phase change, and the reason is that three different fixed costs are crossing their thresholds in the same handful of years. First, specification. Drafting and interpreting the contract that defines a claim used to require a securities lawyer per deal; language models make it nearly free, which collapses the legal and specification line items that gated the long tail. Second, settlement. Programmable money turns the movement of value from a multi-day interbank choreography into a state transition, and pushes compliance — identity, eligibility, reporting — inside the transaction rather than alongside it. Sub-cent payments that were absurd to process are now ordinary. Third, and most important, market-making. A claim with no natural crowd used to be unquotable; a software agent that can research, price, and stand as counterparty to a niche exposure at a fraction of a human's cost makes the long tail economic for the first time.

Any one of these alone moves the boundary a little — the bottleneck simply shifts to whichever cost you didn't fix. It is the simultaneity that matters. When several necessary costs fall together, claims that were each blocked by a different binding constraint cross the line at the same time, and the count of viable markets does not drift upward. It jumps. We are early in that jump, which is exactly why the conclusion sounds implausible: we are reasoning about the far side of a discontinuity from the near side of it.

When fixed costs cross thresholds together, the curve does not slope upward. It jumps.

The mystery is the under-hedging

It is tempting to frame all this as a story about future demand — about exotic new markets people will someday want. That gets it backward. The mystery is not why there will be more hedging tomorrow. The mystery is why there is so little of it today.

A homeowner insures the house but not the income that pays the mortgage. A restaurant that lives and dies by the price of beef and the weather on a Friday hedges neither. A creator whose entire livelihood rides on one platform's ranking algorithm holds that risk completely undiversified. A small business with one big customer, a worker with one employer, a town with one industry — each sits on a large, perfectly nameable exposure and transfers almost none of it. This is Shiller's macro-markets observation in ordinary clothes: we have markets for comparatively narrow financial claims and almost no markets for many of the largest risks people actually bear. If trade in risk creates surplus, then the under-hedged world we actually live in is the anomaly that wants explaining.

The answer is not that these people lack exposures worth transferring. They are drowning in them. The answer is that the cost of naming, verifying, pricing, collateralizing, and settling those exposures has, until now, exceeded the surplus from trading them — so the trades that obviously should happen simply have not. Markets exist because people occupy different positions in the space of risk: different endowments, beliefs, tax situations, horizons, capacities to bear uncertainty. A claim that is a burden to one party is an asset to another. That heterogeneity has been true since the first farmer worried about the price his crop would fetch. What was missing was never the demand.

The demand was always there. The machinery was not.

The future holds trillions of markets

“Trillions” sounds like rhetoric. It is an order-of-magnitude estimate, and the exact exponent is not the point. Reckon the dimensions along which a claim can be defined. The economically relevant entities in one large economy — firms, households, creators, vehicles, buildings, municipalities, machines — already run to the hundreds of millions. Each carries dozens of measurable exposures. Add horizons from minutes to years, locations or contexts, and contract wrappers — equity, debt, revenue share, option, future, swap, insurance, event contract, access right, basket. Even with conservative counts, the product lands between tens and hundreds of trillions of candidate claims before a single basket, spread, or index multiplies the space again.

The right objection is that almost all of those are redundant, uneconomic, or never able to clear the viability bar. Correct — and that is why the conclusion is hard to escape. Discard ninety-nine percent and you are still in the trillions. Discard 99.99 percent and you still clear ten billion: not a precise forecast, but a count unrecognizably larger than today's liquid market surface. The conclusion never rested on optimism about which claims trade. It rests on the dimensionality of measurable economic exposure, and that dimensionality is very high.

Coarse categories give way to continuous prices wherever pricing gets cheap enough to keep. The AAA/AA/A buckets handed down by ratings oligopolies are the credit analog of trading in eighths: low-resolution stand-ins that continuous, machine-priced credit can erode the way the penny eroded the spread.

The number of markets is the pixel count of the economy — and the pixel count is about to explode.

Liquidity is manufactured, not found

Here is the objection a serious reader raises, and it is the right one. Liquidity is finite. Attention is finite. Capital is finite. If every tiny exposure gets its own market, liquidity shatters into a billion thin, manipulable pools and spreads blow out until no one trades. Trillions of markets, the argument runs, cannot all be liquid.

This is decisive against a naive picture of markets and weak against the one actually arriving, because it assumes liquidity is something a crowd must already supply. It isn't. It is something a mechanism can manufacture. Most claims load on a small set of common risk factors — rates, sector demand, energy and freight, regional weather, a credit cycle — plus a residue that is specific to the claim. A market-maker holding a large, diversified book does not need natural two-sided flow in each name. It hedges the factor exposure in deep, liquid markets and warehouses only the idiosyncratic residue, which washes out across many positions by the same law of large numbers that lets an insurer cover a million uncorrelated houses. A dealer can quote a revenue claim on one restaurant by hedging it against a neighborhood-spending index, food-input prices, labor costs, and a basket of comparable restaurants — carrying only the part that is genuinely about that restaurant. This is precisely how options desks already quote thousands of strikes with little flow in any one of them.

The consequence inverts the objection. The number of deep liquidity pools an economy must support scales with the number of hedgeable risk factors far more slowly than with the number of priced claims. Capital and spreads still depend on the residual: covariance, one-sided flow, adverse selection, model error, and stress. But the future does not need trillions of deep pools of human attention. It needs a computational liquidity layer that maps idiosyncratic claims onto a shared, low-dimensional factor structure and warehouses what's left. And for the residue of pure-information claims with nothing to hedge against, a different trick works: a subsidized automated market-maker with a fixed, known maximum loss can quote a continuous price in a market that has no natural counterparty at all.

Fragmentation at the level of claims is consolidation at the level of risk.

What separates a market from a casino

Everyone in finance eventually asks the same blunt question about any new venue: is this risk transfer, price discovery, or just a liquid way to lose money? The framework above answers it, because it turns out there are exactly three ways liquidity gets manufactured, and they are not equally benign.

The first is the dealer who hedges factors: takes on an idiosyncratic claim, lays off its common risk in deep markets, and keeps the residual. The second is the bookmaker — or the insurer — who warehouses a large book of mutually independent bets and lets them net against one another. Here, strikingly, independence is the asset, not the problem: the more uncorrelated the flow, the better it diversifies, which is the exact opposite of the dealer's wish for common factors to offload. The third is the sponsor who simply pays to bring a price into existence, because the price itself is valuable to them even though they capture none of its use.

The bookmaker's regime is the one most associated with “easy and profitable,” and the reason is uncomfortable but clarifying. It runs on uninformed flow. A book full of independent, non-toxic bets is a beautiful business right up until the flow turns informed — which is why bookmakers limit sharps and why the casino, whose customers can carry no private information about a roulette wheel, is the purest and most durable version of the form. That is also the tell for what makes a market good or bad. A market is good when its dominant flow is hedging and immediacy — risk moving from someone who must bear it to someone better able to hold it, both sides ending up better off in the only currency that matters, utility. It can also be good when the price it produces improves outside decisions, even for people who never trade. A market is bad, or at best a wash, when its dominant flow is disagreement-driven speculation with no underlying exposure, no offsetting consumption value, and no decision-improving price: a zero-sum transfer dressed as discovery. Most sports betting is the bad kind. So is the roulette wheel. The mathematics of liquidity does not care about the distinction — a casino can be exquisitely liquid — but the welfare of the people inside it depends on nothing else.

A market is good when it moves a risk to someone better able to bear it, and bad when it manufactures a risk no one needed to hold.

The dual that turns light against itself

An honest version of this thesis has to confront the way its own central force runs in reverse. The story so far treats better information as purely market-creating: cheaper verification lowers cost, sharper signals close the gap that lets the informed fleece the uninformed. But information has a destructive dual, and it bites hardest at exactly the granular, well-predicted, idiosyncratic claims this essay is most excited about.

The point is Hirshleifer's, and it is ruthless. If information arrives that resolves which state will obtain before the parties can trade the claim, the gains from trade evaporate — you cannot insure a house once everyone already knows whether it will burn. The distinction is timing. Verification at settlement creates markets: the model that tells whether the fire occurred after you bought the policy makes the contract cheap to write. Revelation before contracting destroys them: the model that tells everyone whose house will burn before anyone can buy insurance collapses the pool. An AI that prices your individual fire risk to three decimal places does not deepen your fire insurance. Past a point it destroys it, because the premium converges to the now-known loss and the entire rationale for pooling — that none of us knows which of us will be unlucky — dissolves. Perfect foresight is not a complete market; it is no market, because risk requires the very ignorance that the technology is busy eliminating. The same engine that makes a trillion markets possible by making the world legible can, pushed far enough, extinguish the uncertainty those markets exist to trade.

Adverse selection is the other limit, and it is the reason the long tail will not fill in evenly. Whoever sells a claim usually knows more about it than whoever buys, and when the informed side chooses what to offer and when, the pool skews toward the claims the seller is least sorry to part with. Pushed far enough, the discount this forces drives the good claims out of the market and the market with them — Akerlof's lemons, and the single most common reason a market that obviously “should” exist does not. What saves it is verification, and the deep point is that the same infrastructure lowering the cost of building a market is what shrinks this informational wedge: the term and the cost fall together because they are the same technology. So the trillion-market world is not uniform. It is dense along the axis of what can be cheaply verified — cash flows, objective triggers, postable collateral — and sparse off it, in the country of soft information, pure reputation, and the one-shot deal. “Trade my reputation” remains, deservedly, the hardest case.

The force we are betting on can also burn the village it illuminates.

Finance becomes a protocol layer

We treat prediction markets, insurance, and derivatives as three different animals: one trades beliefs, one transfers losses, one moves state-contingent payoffs. They are one animal. An event contract pays out on whether a proposition resolves true; an insurance policy pays a loss subject to a deductible and limit; a derivative pays a function of some future price. All three are the same object — a payoff that depends on the state of the world — and what separates them is regulation, accounting, and collateral convention, not mathematics. It follows that any infrastructure cheap enough to specify, verify, collateralize, and settle one of them expands all three at once.

As the categories converge, finance stops looking like a sector and starts looking like a protocol layer for economic life — the plumbing for moving value across time, risk, and state of the world. The boundary between a “financial transaction” and an ordinary one blurs, because both are transfers of state-contingent value, and the cost of running one has fallen to where it can sit beneath almost any interaction. Seen from that layer, the familiar furniture of the economy reveals its financial shadow. A job is a stream of income. A commute is a portfolio of delays. A song is a stream of royalties. A warehouse is an options book on demand. A reputation is collateral; a forecast is inventory; a favor is a microcontract; a risk is an asset with a negative sign. The future does not invent these exposures. It makes the implicit explicit — turns whispers into bids and inconvenience into price.

A home is a levered claim on a school district, a fault line, and a fire season, with bedrooms attached.

The same wiring carries the cascade

A thesis that narrated only the upside would not deserve to be believed. The mechanism is neutral; its consequences are not, and three of them are worth dreading in proportion to how much one believes the rest of this essay.

The resolution metaphor is a useful lie until it breaks. It makes the dimensionality legible, but prices are not passive pixels on a photograph of the economy. At high enough density, the camera becomes an engine: measuring, ranking, collateralizing, and trading a thing changes the thing being measured. The first harm is predation. Where humans negotiated too slowly to prey on each other at scale, agents do not. In decentralized markets, automated strategies already detect a pending transaction and reorder the queue to skim the value out of it — a class of extraction crypto people aptly call the dark forest — running into the billions a year. In a world where agents negotiate lane changes and crop futures alike, that predation generalizes, and it lands back on the core inequality: a tiny trade with real surplus is destroyed if the expected skim exceeds it. The defense is design — clear orders in discrete batches at a uniform price and the value of being microseconds faster evaporates — but the arms race between extraction and defense is a permanent feature, not a transient bug.

The second is contagion. A world of trillions of cross-collateralized claims is densely wired, and dense wiring transmits shocks. The very architecture that makes the system possible — hedging idiosyncratic claims onto common factors — means a blow to one factor hits everything that loads on it at once, and cross-collateralized players forced to liquidate carry a local failure across nominally unrelated markets. There is a darker version still: if all the idiosyncratic residue gets warehoused by a handful of factor-liquidity engines, the system has dispersed everyday risk while quietly concentrating tail risk onto a few balance sheets. That is the structure of 2008, rebuilt at the scale of everything.

The third harm is human rather than systemic. An economy in which every action carries a price demands relentless calculation, and there is real value in the things that are not for sale — in courtesies with no meter running, in relationships not continuously marked to market. The cautionary artifact here is not a model but a man: years before any of this infrastructure existed, an individual issued a hundred thousand shares in himself and sold his shareholders votes over his own life — what to wear, which party to register with, whom to date. The lesson was not that the market failed but that it worked. His investors approved some life choices, constrained others, and eventually a relationship itself had to be unwound through the shareholder apparatus, with the partner's stake bought out like a position. That a thing can be priced is not an argument that it should be. As markets become ambient, the locus of oversight has to move with them — from the regulation of venues, which assumes markets live in identifiable rooms, to the regulation of claims, which asks of any state-contingent transfer whether it is consensual, capped, disclosed, manipulation-resistant, and free of systemic externality. The room is gone. The questions remain.

Flags planted in the ground

A vision is cheap unless it is falsifiable, so here are the specific, dated calls the argument commits me to. The dates are deliberately staggered: some should be visible within a year or two, others require the full cycle. Some of them will be wrong; that is the price of saying anything worth remembering. If most of them are right by the early 2030s, the thesis was not hype.

1. By 2027

   An agent-to-agent payment or settlement standard clears genuine sub-cent, machine-to-machine transactions in production, with meaningful volume from real services rather than speculation.

2. By 2028

   A prediction-market price is cited on the record in an official government, central-bank, regulator, or Fortune 500 procurement or risk decision.

3. By 2028

   Parametric cover is written and traded at building-and-season granularity — a specific structure, a specific fire or flood window — priced by model and settled by oracle, not by an adjuster's visit.

4. By 2029

   A regulated venue offers a real secondary market in standardized fractional revenue claims on private companies, underwritten directly from live accounting and bank-feed data rather than from a quarterly PDF.

5. By 2030

   Compute and energy trade on standardized, exchange-listed forward and option contracts — inference capacity, latency, and megawatts quoted with enough liquidity for non-specialists to hedge them.

6. By 2030

   A regulated secondary market quotes at least $10 billion of annual trading volume in small-business revenue-share claims against baskets of comparable firms, not just bilateral revenue-based financing.

7. By 2031

   The first synthetically unbundled single-segment equity exposure trades — cloud separated from retail, a “segment swap” — even though the underlying conglomerate never formally splits.

8. Before 2032

   At least one systemic scare is traced to correlated forced liquidation across markets that looked independent — the first cascade through the new wiring. I would rather be wrong about this one.

Own the scarce input, not the venue

If the argument holds, the strategic implication is direct, and it is not where intuition points. The naive move is to build the next venue — another exchange, another listing, another order book. But a venue captures the surplus of the rung it sits on, and in a world of trillions of markets the marginal venue is worth almost nothing. Value accrues instead to whatever lowers a cost term across many markets at once, because that is what brings new markets into existence in the first place.

State it as a law. As claim listing commoditizes, much of the surplus is competed away to users. The capturable rent migrates to scarce complementary inputs: verification, trusted agency, regulatory permission, collateral, balance sheet, execution, and the residual-risk engine. In the old world the scarce input was often the venue, because liquidity was concentrated in venues. In the new world venues commoditize, and the candidate chokepoints move upstream to verification — the infrastructure that turns messy reality into trustworthy, adjudicable inputs, which is simply the new underwriting; sideways to the agents that hold the user's trust and therefore sit at the chokepoint of enormous volume; and inward to the risk engine that prices and warehouses what is left after common factors are netted out. The risk engine is the hardest of these to commoditize. The factors can be named by many firms; the scarce object is the proprietary comparables, residual data, balance sheet, and execution machinery to price and warehouse what remains. That looks less like an exchange than a quantitative market maker for the entire long tail.

For the investor, the surface has two faces and both are the residual seen from opposite ends. The alpha is specialization — knowing the idiosyncratic part that the commoditized liquidity layer leaves unpriced and that adverse selection bites on; edge, in this world, is verifiable knowledge of the specific. The risk is its mirror: basis, model error, oracle failure, liquidity mirages, the cross-collateral cascade. The question stops being “do I like this asset” and becomes “which hidden factors am I warehousing, how crowded is the hedge, and which oracle can break me.” For the founder, the financing stack simply widens — finance a product line without selling equity, hedge your compute and component costs, insure your customer concentration, open an employee-liquidity pool long before any exit. The firm of the future is less a fortress of held assets than a continuously rebalanced portfolio of bought and sold exposures. Which is, in the end, what a quantitative hedge fund already is. The thesis of this essay is that the rest of the economy is about to look like one.

Own the scarce input, not the venue.

Waiting to be priced

We began with a room, and we should end by admitting the room is not coming back. The market was somewhere you went — a pit, a floor, a terminal, an app — because the procedure it ran was expensive enough to concentrate in one place. Drive the cost of that procedure toward zero and it stops being a destination and becomes a medium. It does not open and close; it breathes. It is not a place you visit but a layer you live inside, mostly silent, transacting on your behalf in the background according to standing preferences you set once and revisit rarely: save me money; preserve my privacy; never sell my location; hedge my mortgage; insure my income; sell my flexibility when the price is high; pay for urgency only when it truly matters; refuse markets in the things that should not be for sale.

In that world people will not “trade” the way traders once did, hunched over screens. Their agents will run continuously, and the visible surface of finance will recede into infrastructure the way electricity and packet-switching did — everywhere, noticed almost nowhere. A market for a company. A market for a song's next million streams. A market for a lane change between two cars' agents. A market for rainfall over one valley. A market for a dinner rush at a taco truck on one corner of one city. Trillions of them, flickering in and out of existence as needs arise, some lasting a century and some a second.

The astonishing thing is not that these markets will exist. It is that, in the only sense that matters to the incentives, they already do — here now, in potential, trapped inside illiquidity, opacity, legal friction, and unpriced risk. Every private company already has cash flows. Every small business already has local demand. Every creator already has uncertain earnings; every household a thousand risks; every city scarce space it allocates badly; every person a constant traffic of implicit trades among money, convenience, privacy, attention, and time. None of it is waiting to be invented.

It is waiting to be priced. The natural scale of markets is the natural scale of measurable economic exposure, and that scale is not small. It is combinatorial. It is trillions and trillions.