May 23, 2026

Chokepoints in the AI Hardware Supply Chain

Fifteen layers deep on the thin points that gate the whole compute buildout.

marketsresearch

The thesis

A chokepoint is a single thin point in a supply chain that gates a much larger industry downstream. High-bandwidth memory is the canonical example: constrain HBM supply and you constrain every AI training run downstream of it, no matter how much capital is pointed at the problem.

The interesting property is asymmetry. The chokepoint is usually far smaller than the thing it gates — which means it’s mispriced more often than the headline names, and it moves on different news.

How it’s mapped

Fifteen layers, each one its own report with its own sources and its own charts, built out systematically rather than opportunistically. Each chokepoint gets tracked on:

• Why supply is thin — the actual physical or contractual reason, not a narrative
• What it gates — the downstream industries that stop without it
• Who sits at each layer — upstream inputs, midstream fabricators, downstream integrators, named
• Geopolitical pressure and supply-demand pressure — scored separately, because they move independently
• Catalysts — what would tighten it further
• Disconfirmers — what would loosen it, or kill the thesis outright

That last field is the one that earns its keep. It’s easy to fall in love with a constraint story; writing down in advance what would prove it wrong is the only reliable defense.

Status

The research spine exists — layer reports, a live tracker, a reference library of primary sources. The schema is deliberately built so it can fold into Atlas later: folders map to future tables, fields map to future columns, and every entity has a stable identifier that never gets renamed.