What follows from the map

Six claims, each with the observation that would show it to be wrong.

These are the views this research leans on. Each carries a falsifier: the thing that, if observed, would show it to be wrong. A claim without one cannot be scored, and a claim that cannot be scored is a slogan.

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.

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 show this is 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.
Tested by 1 dated call: Through 2026, high-bandwidth memory supply rather than leading-edge wafer supply will be the stated limit on AI accelerator shipments. (resolves by 2027-03-31)

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 covered here.

What would show this is wrong: Sustained periods where chokepoint prices track production cost, with substitution or new entry arriving fast enough to cap them.
Tested by 1 dated call: Silicon carbide substrates will not become a binding constraint before 2028: no sustained shortage, and at least one more Western capacity retrenchment. (resolves by 2028-01-31)

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 show this is wrong: Interconnection queues and transformer lead times shortening while announced datacentre capacity keeps rising — energy ceasing to be the thing that slips.
Tested by 2 dated calls: Large power transformer lead times will still be three years or longer at the end of 2027, despite the capacity expansions announced through 2026. (resolves by 2028-01-31) Median US interconnection queue duration will not fall below three years by the end of 2028, even though FERC Order No. 2023 took effect in November 2023. (resolves by 2029-03-31)

“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 show this is wrong: Qualification cycles compressing materially — second sources reaching production in quarters rather than years.
Tested by 2 dated calls: Neon will not see another supply squeeze of the 2022 kind before 2028, because on-site recycling has structurally cut what a fab needs. (resolves by 2028-01-31) The US Section 232 action on polysilicon will not add qualified electronic-grade capacity in the United States before 2029, whatever it does to solar-grade supply. (resolves by 2029-06-30)

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 show this is wrong: A durable de-escalation in which controls are lifted and re-shoring programmes deliver qualified capacity at scale.
Tested by 1 dated call: China will still require export licences for dysprosium and terbium at the end of 2027, and no non-Chinese source will supply more than a tenth of world heavy rare-earth separation capacity. (resolves by 2028-03-31)

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