What it is: build the bill of materials for a thing, then run it backwards off mandatory regulatory filings to find what it actually costs and who actually captures the margin.
When to reach for it
When a company's story and a company's economics might not match — and there is a filing that has to be true. Reach for it when the number everyone repeats came from a press release.
The steps
- Build the forward BOM. What is this thing physically made of? Major subsystems, then parts.
- Name the supplier for each line. Who actually makes it. This is where concentration shows up.
- Go to the filings, not the feed. SEC EDGAR full-text search — 10-Q, 10-K, S-1. His words: "filings that are mandatory and are due to scrutiny."
- Run it backwards. Segment revenue and cost of revenue against your BOM lines. The gap between what a part costs and what the system sells for is the answer.
- Ask who is a single point of failure. If one supplier appears on every line, that is the story.
Worked example
- Applied to NVIDIA (Moonshots #284
0:00) → the panel holds NVIDIA as an index of the era. The method asks a different question: "why should I start looking at financing coming from anywhere except going to NVIDIA? They have a ton of cash, they have the technology, they have the interest." Then it surfaces TSMC as the single point of failure under the whole stack — a conclusion the panel never reaches, arrived at by looking at parts instead of price. - Applied to launch → the idiot index piece, published 2026-07-08, GForce's best-performing post to date. Same method, different object.
Where it fails
- It cannot price private companies. No filing, no method. SpaceX before the S-1 is guesswork and should be said as guesswork.
- Segment reporting is coarser than a BOM. You will often get "Data Center: $89.0B" and no way to split it. Do not manufacture the split.
- It says nothing about demand. A perfect BOM tells you what a thing costs and nothing about
whether anyone will buy it. That is the downstream half → see the
## Downstreamsection on every entity page. - ⚠️ It invites false precision. On tape, "89% versus 44" was not a growth rate. Run
verify-firston every figure the method produces.
Where I learned it
Hardware. You cannot ship a system without knowing what it is made of, and once you have costed a real BOM you stop believing headline numbers about anything.
Reading rule. Anything marked unverified or on-air is a claim made on the podcast and has not been checked against a primary source yet. It stays marked until it has. Sourced numbers link to where they came from.