Position · August 2026
What cannot compress.
Analysis is compressing toward free. The work a strategy firm once sold by the year is already producible in days, and the direction of travel is not in serious dispute. This page is a dated position on what follows: three things cannot compress with it, and capital, wherever it is thinking clearly, is already reorganising around them.
01 · The compression
Argument
When analysis is free, it stops being the product.
Split any advisory business into two components: execution, the analysis, drafting, modelling and assembly that recursive intelligence performs at collapsing cost, and verification, the physical, real-time, non-repeatable calibration of claims against actual outcomes. The two combine as near-perfect complements, which means aggregate value is bottlenecked by whichever is scarcer. Execution is compressing toward free. Verification is not, because reality emits evidence at its own rate and no amount of reasoning accelerates it.
The consequence is already visible in miniature: reports, decks and models, the artefacts an industry priced for a century, are becoming transient projections that anyone can regenerate. What persists is the thing the artefact was projected from. The report is transient. The state estimate persists, and so does everything else a model cannot conjure on demand.
02 · The three scarcities
What survives
Three things do not compress, and each is already priced wrongly.
- Ground truth
The capacity to reason compounds. The rate at which reality emits evidence does not. Observation of the physical world, the joint that held, the asset that ran, the counterparty that performed, arrives in windows that do not reopen, and a record of checked outcomes cannot be compressed or bought at any price once the window has passed. Lead time is the moat, and lead time not started is lost permanently, not deferred.
- Physical position
On our working estimate, compute becomes the binding economic constraint within this decade, and compute is not an abstraction. It is land that is entitled, power that is connected, water that is allocated, equipment that is queued and delivered. Every one of those is fixed in place, granted by authorities on their own clocks, and indifferent to how intelligent the bidder's model is. A queue position cannot be reasoned into existence.
- Standing
Licences, registrations, counterparty files, and the slow records of conduct that let strangers transact quickly. Trust of this kind is written one dated entry at a time and survives audit precisely because it cannot be generated. In a world of free analysis, standing becomes the price of admission, because everyone's arguments are excellent and only some people's records are real.
03 · The consequence for capital
Where value settles
Capital settles around compute and the hard assets either side of it.
Follow the three scarcities to their asset classes and the map draws itself: compute capacity, and the hard infrastructure upstream and downstream of it. Power generation and the grid positions that connect it. Water. Entitled land in the right districts. The shells, cooling and fibre that turn electrons into capacity, and the logistics that feed all of it. We have written elsewhere that in data centres the constraint stack has inverted and land ranks fourth behind power, water and equipment: that page was this position wearing work clothes.
Two corollaries matter for anyone allocating. First, the familiar instinct of private capital toward real assets is not nostalgia; it is rung one of the same ladder, because the substrate of compute is land, power and permission, and the disciplines that govern a building govern a data hall. Second, rent follows ownership, which is mobile, while capacity is fixed in place, and in a contested decade that difference is a hedge: the owner's domicile travels, the asset's jurisdiction does not, and portfolios should be built knowing which of the two they are exposed to.
04 · The position
Dated, falsifiable
Taken in August 2026. Amended in the open, never quietly.
The position: as analysis compresses toward free, durable value concentrates in ground truth, physical position and standing, and the decade's serious capital settles around compute capacity and the hard infrastructure upstream and downstream of it, entering through the real-asset disciplines it already knows.
What would change our mind, stated now so the record can be checked against it. On ground truth: if verification quality keeps improving on purely synthetic or third-party data, with no dependence on accumulated physical observation, the first scarcity fails and we will say so. On physical position: if efficiency gains outrun demand so decisively that compute never binds, the second weakens. On standing: if institutions begin transacting at scale on unverified model output without records of conduct, the third was sentiment, not structure. Each of these is watchable within years, not decades, and this page will carry the result either way.