Reading the envelope: what extraction can afford¶

A product announcement is close to unforecastable. A datacentre is not. One is decided in a quarter by people who could have decided otherwise; the other is a multi-year commitment of capital, land and grid capacity that was visible long before anything was announced.

Reading the envelope makes that trade deliberately, watching slow constraints instead of fast decisions. The constraints move over quarters and years, which makes them legible in advance, and the price of the legibility is precision. Readings of this kind come out directionally right and chronologically early. Anyone wanting a date is better served elsewhere.

The same gauges can be built for the data economy. They describe what collection can afford to do, not what any particular firm will try.

What pays for all of it¶

Behavioural extraction is funded, still, almost entirely by advertising. That is the master gauge, and its important property is that it does not expand because a pipeline would like it to. Advertising has held a fairly stable share of economic activity across decades, through several technology cycles, which puts a ceiling over everything built to serve it. Collection can get more precise, cheaper, and more invasive without the pot getting larger.

Most of the industry’s strategic behaviour over the past decade reads as an attempt to get out from under that ceiling. Retail media networks, subscription tiers, payments, data licensing, and sales into credit, insurance, employment screening and government all move revenue away from the auction and toward buyers with separate budgets. The landscape of commercial extraction describes the resulting sprawl.

The number worth watching is not headline advertising spend. It is the share of a platform’s revenue arriving from anywhere other than advertising, because that share measures how much of the system has found a second funder.

What identifiers are still available¶

The second master gauge is the supply of stable identifiers, and it behaves much like component access does for a war economy: a ceiling that lowers slowly and forces substitution downward rather than a stop.

Cookie deprecation, mobile advertising identifier restrictions and app tracking permissions degraded collection without ending it, and degradation leaves a signature. Deterministic joins give way to probabilistic matching. Hashed email spreads as a fragile deterministic substitute that depends on people using one address. Client-side tags move server-side, where a publisher relays what a browser would have blocked. Clean rooms appear as a contractual workaround for a technical loss, letting two parties compute over data neither can hold. Each step costs accuracy, and each is adopted because the previous rung is gone.

Falling match rates and rising reliance on modelled rather than observed audiences are the readable symptoms. At ground level this is the same gauge a reader touches when they reset an advertising identifier or turn off Topics: the individual gesture is small, and it is small in a direction the whole industry is being pushed.

Whether a second buyer arrives¶

Model training created something the data economy had not previously had, which is a large buyer of behavioural and content archives that is not an advertiser. Whether that demand hardens into a durable revenue line or resolves into a handful of one-off licensing deals is the open question with the widest downstream effect, since a genuine second buyer would fund collection that advertising economics could never justify on its own.

This gauge is faster-moving than the others and borrows its slowness from underneath. The demand appeared quickly; the compute, fabrication capacity and power supporting it are physical commitments measured in years, and those are forecastable even when the demand is not. Where the buildout continues past the point the current revenue supports, somebody is betting on the second buyer being real.

The cost of keeping everything¶

Falling storage cost is the standard explanation for why retention beat deletion, and as a gauge it has gone stale. A cost that only ever falls behaves as a constant, and constants do not predict anything.

The live version is holding cost weighed against the expected value of holding, and recent movement has come from the supply side rather than the demand side: constrained memory and drive supply, grid connection queues, and power as the binding input for anything doing inference at scale. Where retention starts carrying a real marginal cost again, minimisation acquires an ally it has not had for twenty years, which would be an odd and welcome way to win an argument.

What a regulator can actually process¶

Statutes are the wrong thing to count, and second-order effects already sets out how unevenly the written rules land. The forward-reading number is capacity: caseworker headcount, time from complaint to decision, how many large investigations an authority can run at once, and whether penalties are collected or appealed into irrelevance.

Capacity changes over budget cycles, which makes it slow enough to read and slow enough that a sudden statute rarely changes it. The private channel belongs on the same gauge and moves faster: class actions and cyber insurance price risk on a commercial clock, and insurers reshape behaviour through exclusions rather than judgements.

Where the ground itself sits¶

The slowest gauge, and probably the most genuinely predictive. Two mobile operating systems decide what an application is able to observe. A few cloud providers host most of the processing. A small number of exchanges clear most of the demand. Very little in this domain happens without passing through one of them.

Concentration cuts both ways, which is what makes it interesting rather than merely grim. It makes extraction efficient, and it makes regulation tractable, because a rule aimed at four companies can be enforced in a way that a rule aimed at forty thousand cannot. It also means a platform policy change can accomplish in one release what a directive spends six years failing to achieve, and that the same power is available for the opposite purpose. Where that concentration ends up sitting is argued at companies and, where states are the actor, nations and states.

Disposition changes the response, not the price¶

Constraints set the option set. They do not pick from it, and identical conditions produce opposite behaviour depending on who is facing them.

Ownership structure, time horizon and exposure to a consumer brand all modify the response. Founder control through dual-class shares tolerates regulatory risk differently from quarterly earnings exposure, and private equity ownership ahead of a sale differently again. Facing the same identifier scarcity, one firm degrades gracefully into contextual advertising while another escalates into fingerprinting, and the difference is disposition rather than arithmetic. States enter here too, since a government finding the commercial ecosystem a convenient source of its own intelligence is not a neutral party to that ecosystem’s constraints.

This is also why sanctions-style predictions disappoint in both domains. Pressure applied to revenue is tolerated far longer than pressure applied to survival, and the two get measured as though they were the same gauge.

What the envelope will not say¶

It gives no dates. It names no specific move, and it cannot rank which firm goes first. Several of its gauges are measured through figures published by the industry being measured, which is a real weakness rather than a formality. A single ruling or one platform release can outrun every reading on the list.

The gauges were selected for slowness, so the method is blind to fast shocks by construction and will read them as surprises afterwards. That is the cost of the trade, and it is worth paying only for the narrower question the method actually answers: which of the current arrangements are being held up by a constraint that is loosening, and which are quietly becoming unaffordable.

That question is what makes systems effects actionable rather than merely true. The homeostatic trap, a system’s resistance to reforms that would cut the function it is optimised for, explains why the equilibrium holds; the gauges indicate where it is under load. The backward-facing companion, reading a claim against what was already true, is residue.

Last reviewed: 2026-07-19.