Thesis

What we think follows from the map — stated so it can be scored, not admired.

A research house is supposed to have a view and be judged on it. Each claim below carries a falsifier: the thing that, if observed, would show it to be wrong. A thesis without one is a slogan, and a slogan cannot be scored.

This is a general view published to whoever reads it. It is not advice, it is not addressed to anyone in particular, and it takes no account of your circumstances — see Acting on it for what that does and does not permit.

The bottleneck is rarely where the attention is.

Attention, and therefore price, concentrates on the visible layer — the model, the accelerator, the hyperscaler. The binding constraint usually sits one or two layers below it, in a company nobody writes about. The gap between where a chain is priced and where it actually binds is the whole reason to map it.

What would prove this wrong: If constraints resolved at the visible layer — if compute output tracked chip design announcements rather than packaging, power and tooling — the map would be describing a chain that no longer gates anything.

A chokepoint does not price like a commodity.

When both demand and supply are inelastic, price is set by scarcity rather than by cost of production, and it moves in jumps. Cost-plus intuition — the instinct that says a material cannot be worth many times its input cost — systematically misprices exactly the nodes this firm covers.

What would prove this wrong: Sustained periods where chokepoint prices track production cost, with substitution or new entry arriving fast enough to cap them.

This decade, energy binds before silicon does.

Compute is being announced faster than it can be energised. Transformer order books, interconnection queues and turbine slots clear on a slower clock than fab construction, and none of them can be accelerated with capital alone.

What would prove this wrong: Interconnection queues and transformer lead times shortening while announced datacentre capacity keeps rising — energy ceasing to be the thing that slips.

“There is an alternative” is usually false on the horizon that matters.

Qualification is measured in years: a second-source material, tool or resist has to be proven per process, per fab, per application. A substitute that exists in a laboratory and a substitute that is qualified are different facts, and only the second one relieves a constraint.

What would prove this wrong: Qualification cycles compressing materially — second sources reaching production in quarters rather than years.

Supply concentration is now a policy variable, in both directions.

Export controls made geography a first-order term in these chains. That cuts both ways: restriction raises the value of what is restricted, and subsidised re-shoring can destroy the scarcity that made a node interesting in the first place.

What would prove this wrong: A durable de-escalation in which controls are lifted and re-shoring programmes deliver qualified capacity at scale.

The map compounds. A position does not.

Any single view can be wrong and is eventually closed. Knowing every qualified producer of a material, and being told when a row is wrong by someone who works in that chain, is an asset that accumulates. That is why the research is published rather than sold, and why it comes before any book.

What would prove this wrong: The map failing to attract corrections — no practitioner engagement — which would mean it is not compounding, only ageing.