Actionable insights — Netting out circular revenue
The repeatable analysis behind the call: not what to sell, but how the revenue line is tested — written so the process can be rerun later on different names.
Scope note. Only the free preview of this paid post was capturable, so these are the methods visible in its opening section. The paywalled remainder ("Looking Closer at NVIDIA's Numbers" onward) certainly contains more, and none of it is represented here.
How to read this page: the argument in the captured section is not a valuation argument — no multiple, no target, no price is mentioned. It is an argument about revenue quality, and the technique behind it is portable to any capex cycle where the vendor also finances the customer. Each insight is the adjustment to make before trusting a reported number, and the signal to watch when re-running it. The boxed line shows how it played out here.
1. Net the seller's own funding out of the seller's revenue before believing the growth
The repeatable method
- Take the headline revenue line for the new business the story rests on (here, "AI revenue").
- Identify every customer inside it that the seller is also an investor in, lender to, or compute-credit provider for. Related-party and equity-method disclosures, plus press coverage of investment rounds, are enough to build this list.
- Subtract those customers' contribution. Call what remains the arm's-length line.
- Restate the growth rate on the arm's-length line only. That is the number that tests whether independent demand exists; the headline tests only whether the financing arrangement is still running.
- Repeat one link up the chain: the seller's own suppliers are exposed to the same arm's-length figure, not to the headline.
Here:
MSFT reported $34.3bn of fiscal-2026 AI revenue; $24.1bn came from OpenAI alone, a company Microsoft funds. The arm's-length line is "a single-digit-billion-dollar business" — roughly $10bn. The same adjustment applied one link up hits NVDA, whose demand is the hyperscalers' >$1tn of GPU capex.
Watch for
- Disclosure changes: a vendor that stops breaking out the customer, or folds the AI line into a broader segment, is removing the input this test needs.
- The funding cadence, not the revenue cadence — new investment rounds into the customer are the leading indicator of the next few quarters of "revenue."
- Non-cash consideration: compute credits, equity taken in lieu of payment, and prepayments funded by the vendor all belong in the subtraction.
2. Read customer concentration forward, not backward — and count the pair, not the name
The repeatable method
- For the division carrying the story (a cloud segment, a chip line), find the projected share of revenue attributable to the top one or two customers next year, not the historical share.
- Group counterparties that share a common dependency into a single exposure — two labs funded by the same set of backers are one risk, not two.
- Check the counterparties' own solvency. Concentration in profitable, self-funding customers is a commercial risk; concentration in loss-making, externally funded ones is a financing risk, which turns much faster.
- Size the drawdown as if the whole grouped exposure reprices at once, because a common funding source means it will.
Here:
OpenAI and Anthropic are estimated at ~70% of hyperscaler AI revenue, and UBS projects them at 48% of all Google Cloud revenue next year — so GOOGL's cloud growth is being underwritten by two private counterparties that "remain deeply unprofitable, having lost tens of billions of dollars between them."
Watch for
- The first quarter in which a hyperscaler's AI revenue decelerates while its capex guidance does not — the loop tightening before it is acknowledged.
- Any disclosure from either lab (an S-1, a funding round's terms) that puts an auditable number on losses and on who is paying whom.
- A third-party customer base actually emerging: broad enterprise AI spend that is not routed through the two labs would break this entire objection.
Methods distilled from the publicly visible free preview of a paid Substack post. Not investment advice.