ED ZITRON: AI revenue is circular: the same money shows up as investment, cloud revenue and lab revenue

The Gist

Ed Zitron says big tech companies give AI labs money, the labs spend it on those companies' clouds, and both sides then report it as growth. A few labs make up a large share of their suppliers' revenue, so the growth looks bigger than real outside demand. This is a steelman: Ed Zitron's arguments from the conversation are reconstructed in their strongest fair form, with his answers to the host built in. It does not endorse them.

Conclusion

Much of the AI-related growth at hyperscalers and suppliers traces back to a few labs funded by those same companies, so headline revenue growth overstates independent demand.

Premises

1. Amazon and Google invest in AI labs: Google committed $10 billion to Anthropic with up to $30 billion more, Amazon put $5 billion into Anthropic with more financing available, and Amazon committed $50 billion to OpenAI.
2. The labs spend heavily on those same clouds: Anthropic has minimum spend commitments of at least $110 billion with Amazon and about $114.5 billion with Google.
3. About 48 percent of Anthropic's first-half 2026 revenue came through cloud marketplaces such as Amazon Bedrock and Google's platform; Anthropic books those sales at their full amount, and the same sales also run through the clouds' own businesses.
4. Cloud is the main growth segment in the hyperscalers' recent earnings.
5. Other suppliers show the same concentration: OpenAI was about 11 percent of Microsoft's quarterly revenue, and a single customer reported to be Anthropic was 34.5 percent of SpaceX's Q3 revenue.

Assumptions

• Disclosure: Ed Zitron is CEO of EZPR, a media relations and public relations firm. He writes the Where's Your Ed At newsletter, which has a paid premium tier, and hosts the Better Offline podcast, produced with iHeartRadio and Cool Zone Media; both cover the AI industry critically, and in this conversation he says he predicted the rating outcome ("told you so"). Sources: "About Where's Your Ed At" (Where's Your Ed At) at https://www.wheresyoured.at/about/, "Introducing: Better Offline" (Where's Your Ed At) at https://www.wheresyoured.at/introducing-better-offline/ and "Ed Zitron" author page…
• The conversation was recorded on October 30, 2026, as the FT story on Anthropic's expected rating broke; figures are as stated live and may be approximate.
• "Circular" means money that a company invests in a customer coming back to it as that customer's purchases.
• Differs: Anthropic books the cloud provider fee as an operating expense, and the clouds act as collection agents; public filings do not show the same revenue booked in full by both sides.
• Differs: SpaceX's filing does not name Customer B; the Anthropic identification rests on an anonymous source and has not been confirmed by either company. Microsoft's figure also includes revenue-sharing payments.

Analysis

Argument type: Inductive.

Potential Fallacies

• Equivocation on 'circular' (flow of funds vs. double-counting) (Title, P3, and the conclusion's 'overstates') — The title and P3 suggest the same revenue is counted several times. Under the stated assumptions, circularity only means invested money returning as purchases, and filings do not show duplicate full booking. The evidence supports the weaker flow-of-funds claim, not an accounting inflation claim.
• Non sequitur / missing bridging premise (P1–P5 to the conclusion; P4 specifically) — Investment ties, spend commitments and customer concentration are treated as showing that demand is not independent. The argument does not show that the labs' purchases are financed mainly by the investors' capital rather than by end-customer revenue or other funders, or that lab spend is the main driver of cloud growth (P4).
• Concentration conflated with circularity (P5) — A large customer is not necessarily a customer funded by its supplier. The SpaceX example has no stated investment link to Anthropic, and the Microsoft figure includes revenue-sharing, so these measure dependence rather than round-tripping.
• Commitments treated as realized flows (P1, P2) — 'Up to', 'more available' and 'minimum spend commitments' are optional or forward-looking obligations spread over several years. They are not disbursed investment or recognized revenue, and they depend on the labs' continued funding.
• Hasty generalization / missing denominator (P3–P5 to the conclusion) — Examples centered on Anthropic and OpenAI, plus two heterogeneous supplier data points, are generalized to 'much of' AI growth at hyperscalers and suppliers without giving the labs' share of total growth. Large absolute numbers invite inferences about proportion without context.

Counterarguments

These challenges are not ranked.

• Conclusion (High impact) — The labs sell real products to paying enterprises, developers and consumers. Cloud revenue from lab compute therefore largely reflects independent end-customer demand routed through a shared channel. Investor funding changes the capital structure, not the authenticity of downstream demand.
• Premise 1 and Premise 2 (High impact) — The minimum commitments (~$225B combined) dwarf the investments (tens of billions), so most of the labs' spend must be financed by external customers and other investors. The returning dollars are therefore not simply the invested dollars, and the strict circularity definition applies only to a fraction of the flow.
• Premise 3 (High impact) — Under A4 the cloud fee is an operating expense for Anthropic and the clouds act as collection agents, so filings do not show duplicate full booking. The strong 'same money counted twice' framing is not supported, and gross-versus-net presentation is common to all marketplaces.
• Premise 4 (High impact) — Hyperscaler cloud growth may be driven substantially by non-lab AI workloads, traditional migrations and pricing. If so, the lab-linked, investor-funded share of growth is small and the overstatement is immaterial.
• Premise 5 (Medium impact) — Customer concentration is normal in early infrastructure buildouts and is a disclosed risk factor rather than evidence of fabricated demand. The SpaceX identification is unconfirmed, and the Microsoft figure includes revenue-sharing, so these examples do not demonstrate a circular pattern.
• Conclusion (Medium impact) — The reasoning proves too much. If supplier investment in a customer makes that customer's purchases non-independent, then Nvidia's investments in its customers, corporate venture capital and other vendor financing would all be suspect. The argument supplies no threshold separating ordinary strategic financing from distortion.
• Conclusion (Low impact) — Investors also earn equity upside and mark-to-market gains on their stakes, so the economics are not a zero-sum loop. The investments are strategic, securing frontier-model access and long-term compute demand, rather than a device for manufacturing revenue.

Suggested Improvements

• Quantify the lab-attributable share of growth — Use segment disclosures, backlog or remaining-performance-obligation data and earnings commentary to estimate what fraction of AWS, Azure and Google Cloud growth comes from the funded labs, and state a threshold for 'much'. The conclusion depends on this numerator and denominator, and neither is currently provided.
• Trace the source of lab funds — Estimate what share of the labs' compute spending is funded by investor capital versus customer revenue and third-party funding rounds, and net investment against spend over a matched time horizon. This determines how circular the loop actually is under the stated definition and tests the counterfactual that spend would have occurred anyway.
• Reframe the claim to match the evidence — Make the central claim one of concentration, related-party dependence and correlated counterparty risk, rather than double-counted or inflated revenue, unless a stronger accounting case is made. The stated assumptions concede no duplicate full booking. A claim matched to the evidence is harder to rebut.
• Separate commitments from realized flows — Distinguish firm, funded, cash investment from 'up to' tranches, and contractual minimums from recognized revenue. State time horizons and termination or take-or-pay terms. Timing and contingency determine whether the commitments show current-period inflation or ordinary long-term procurement.
• Strengthen sourcing — Tie each figure to primary filings or company announcements, drop or flag the unconfirmed SpaceX identification, and adjust the Microsoft figure for revenue-sharing. The current evidence rests largely on one live, approximate, and interested source, so primary corroboration would raise reliability.
• Add a baseline and a limiting principle — Compare against normal vendor-financing and customer-concentration levels, historical cases (such as telecom vendor financing circa 2000), and a test that separates benign strategic investment from distortive round-tripping. Without a benchmark, 'circular' and 'overstated' carry alarm without a measurable standard.

Scenario Tests

• Anthropic's credit rating comes in weak and its financing tightens, forcing it to cut or renegotiate cloud commitments. (Supports) — Cloud backlog and growth would be shown to depend on lab solvency, confirming correlated counterparty risk. It would confirm dependence more than prior double-counting.
• Disclosure shows that the funded labs account for a small fraction of AWS and Google Cloud growth, with most coming from non-lab enterprise and traditional workloads. (Challenges) — The conclusion's 'much of the growth' claim would fail on magnitude, even though the investor-customer loops would still exist.
• Most of Anthropic's marketplace revenue is shown to come from third-party enterprises that would buy Claude regardless of who invested in the company. (Challenges) — The marketplace revenue is genuine independent demand routed through a shared channel, and circularity would apply only to the portion of compute funded by investor capital.
• A lab's spending is shown to be funded mainly by hyperscaler tranches, with little outside revenue or capital, as with a lab whose revenue is largely cloud credits. (Supports) — Under the stated definition the demand is substantially circular, and headline growth would overstate independent demand for that portion.
• The SpaceX Customer B is confirmed to be Anthropic, with no investment link between SpaceX and Anthropic. (Neutral) — It would confirm concentration risk among suppliers. But without supplier financing it would not demonstrate circularity under A3, so the conclusion would gain little.
• A reader applies the same 'funded customer means non-independent demand' test to Nvidia's investments in its customers and to corporate venture arrangements generally. (Challenges) — The test condemns most strategic financing without a materiality threshold, which indicates the argument needs a limiting principle.

Coherence & Relevance

The argument is coherent as a narrative: investors fund labs, labs commit spending back to the same clouds, and the resulting revenue is concentrated. That narrative supports a claim of entanglement, concentration and counterparty risk. It does not support the stronger conclusion that headline growth overstates independent demand, because the argument never quantifies the labs' share of growth or the proportion of lab spending funded by the investors' own capital. The stated assumptions also narrow the claim, conceding no duplicate booking and an unconfirmed customer identity. The title's wording therefore outruns the evidence, and the more defensible conclusion is a directional warning about dependence rather than a measured overstatement.

• Amazon and Google invest in AI labs: Google committed $10 billion to Anthropic with up to $30 billion more, Amazon put $5 billion into Anthropic with more financing available, and Amazon committed $50 billion to OpenAI. (Strong) — Establishes the investor-customer link, but not how much of the invested money is actually funded, or that the investment drives the purchases.
• The labs spend heavily on those same clouds: Anthropic has minimum spend commitments of at least $110 billion with Amazon and about $114.5 billion with Google. (Strong) — Closes the loop, but commitments are multi-year, non-realized and far larger than the investments. The share funded by investor capital, and any time horizon, are missing.
• About 48 percent of Anthropic's first-half 2026 revenue came through cloud marketplaces such as Amazon Bedrock and Google's platform; Anthropic books those sales at their full amount, and the same sales also run through the clouds' own businesses. (Moderate) — Shows entanglement of lab and cloud revenue streams, but not that the demand is non-independent. It is a share of the lab's revenue, not of cloud growth, and A4 undercuts the double-counting implication.
• Cloud is the main growth segment in the hyperscalers' recent earnings. (Moderate) — It is the pivotal link from lab flows to headline growth, but it is unsourced and has no decomposition of growth by customer. It does not show that lab spend drives the growth.
• Other suppliers show the same concentration: OpenAI was about 11 percent of Microsoft's quarterly revenue, and a single customer reported to be Anthropic was 34.5 percent of SpaceX's Q3 revenue. (Weak) — Concentration is not circularity. The SpaceX identification is unconfirmed with no stated investment link, and the Microsoft figure includes revenue-sharing. The two cases are heterogeneous and do not generalize to suppliers broadly.

View this argument on LogicFirst.ai