Episode 8 · 11 Aug 2026

Chips, clouds, and who actually gets paid

Following the money through the AI supply chain: silicon, cloud capacity, model providers and the applications on top, and why the order matters more than the headlines.

Show notes

• In a gold rush, map who sells the picks before celebrating the miners. The layer that collects rent is rarely the layer that gets the coverage.
• Capital intensity is a moat only if you own it. Renting your moat by the hour is a cost line, not a strategy.
• When an application company's largest expense is someone else's product, the pricing power sits elsewhere.

Chapters

• The week in five lines
• The supply chain, layer by layer
• Where the margin lives
• Listener mail

Transcript

This episode runs as written notes rather than a recording. The argument, in order.

Start at the bottom of the stack and work up. Silicon is scarce and expensive, so silicon gets paid first. Cloud capacity is bought in bulk and resold by the hour, so the cloud gets paid second. Model providers sit in the middle, spending heavily on both sides and betting that scale becomes pricing power later. Application companies sit on top, and their single largest cost is frequently the model underneath them.

That order is not a prediction about who deserves to win. It is a description of where the cash currently goes, and it matters because coverage of the industry tends to invert it. The loudest announcements come from the layers with the thinnest margins.

The practical takeaway for buyers and founders is the same: know which layer your supplier lives in, and know whether they own their cost base or rent it. A company that rents its core capability by the hour will eventually pass the rent to you.