ED ZITRON: The AI boom is an investment operation, not a business: revenue is circular and venture-funded, the wider economy shows no payoff, and the money will run out

The Gist

Ed Zitron says the AI industry mostly runs on investors' money moving in a circle, not on customers getting real value. Normal businesses are not seeing gains, borrowing is getting harder, and when the money runs out the whole thing shrinks. 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

Because AI revenue depends on circular investment and venture-funded demand rather than returns earned in the wider economy, and because the financing behind it is getting dearer, as Anthropic's expected rating shows, the industry is built on resources that will run out, and when they do, its revenue and spending will fall.

Premises

1. The expected junk rating means Anthropic will finance its commitments at higher cost from a smaller pool of lenders, with more risk on the suppliers that back it, so its compute commitments weigh more heavily as liabilities, and OpenAI faces the same problem in a worse form.
   • Needed for the conclusion
   • Unresolved objection to this statement
   • The link from a lower rating to dearer, narrower funding is standard credit logic, although the rating is only expected and its size is unquantified.
2. So far the evidence shows AI paying off mainly for the companies that sell it, while at Google it is slowing the core business it was meant to strengthen.
   • Needed for the conclusion
   • The payoff claim and the Google claim come with no data or baseline, and lags in diffusion offer an equally natural explanation.
3. 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.
   • Needed for the conclusion
   • The reliance of supplier growth on a few funded labs is a checkable structural fact. How much of it is circular is left unquantified.
4. A large share of AI lab revenue rests on demand subsidized by investors rather than earned through returns, so it cannot be repeated in a normal economy; when venture money and cheap debt tighten, revenue will fall, which makes the industry an investment operation more than a business.
   • Needed for the conclusion
   • Unresolved objection: A substantial base of enterprise and developer customers who pay from operating budgets, with falling serving costs, would mean much of the revenue survives tighter money. The argument does not address how large this base is.
   • The size of the subsidized share and customers' response to higher prices are neither measured nor argued, yet the revenue-decline step depends on them.

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.
• The argument predicts fragility and an eventual contraction; it does not give a date.

Analysis

Reasoning breaks down

Never shows how much AI revenue would vanish without subsidy

The argument gives good reason to think the labs' funding is getting costlier and riskier, and that money loops among a few players. But to conclude that revenue and spending will fall, it needs most demand to exist only because investors are covering part of the cost. It does not say how big that share is, and it does not show that customers would stop buying at higher prices. Without that link, the same facts fit a risky, debt-heavy growth story as well as a coming decline.

Why

If revenue breakdowns showed that most of it came from customers who stop buying when prices rise or when funding dries up, the conclusion would follow.

Argument type: Inductive.

Potential Fallacies

• Hasty generalization (Premise 1) — An expected rating for one lab, plus the claim that a second lab is worse off, is stretched to the whole industry's funding. Hyperscalers and other participants have very different funding bases, including large operating cash flows, so one lab's credit position does not transmit to them directly.
• Treating a financing constraint as a demand constraint (Premise 4) — Costlier credit raises the cost of supply and of commitments. Revenue falls only if customers are buying because of below-cost pricing and labs cannot reprice, cut costs or find other backers. The argument slides from a margin or solvency problem to a demand problem.
• False dichotomy / loaded definition (Premise 4) — The contrast between an investment operation and a business treats subsidized growth and real commercial demand as mutually exclusive. Many capital-heavy firms were both for years, so the label carries more weight than the evidence gives it.
• Appeal to ignorance (absence of evidence) (Premise 2) — A lack of visible economy-wide payoff so far is read as evidence that none will appear. General-purpose technologies usually show long adoption and measurement lags, so the observation window may be too short to settle it.
• Unfalsifiable prediction (Conclusion) — An undated claim of eventual contraction in a cyclical, capital-intensive industry will almost certainly be true at some point. Continued growth can always be read as 'not yet', so confirmation adds little information about the causal story.
• Partial circularity / begging the question (Premise 4) — P4 asserts dependence on subsidy and future decline, which is close to restating the conclusion. The crucial step is assumed rather than derived from the more checkable financing and concentration facts.

Counterarguments

The step from costlier financing to falling revenue depends entirely on a large, non-durable subsidized share of demand. That share is never measured, and the argument does not address how customers would respond to higher prices.

Without this step, nothing else in the argument reaches the conclusion.

• Strongest challenge: Premise 4 (High impact) — Much lab revenue comes from enterprises, developers and consumers who pay because the tools are useful, and falling inference costs shrink any subsidy gap. Customers buying a discounted product may keep buying at higher prices. Tighter financing would then slow growth and squeeze margins rather than cut revenue. Many subsidy-built businesses became durable.
• Argument as a whole (High impact) — A junk rating for one leveraged lab shows that lab is financed aggressively, not that the industry's funding is finite. Hyperscalers fund most spending from operating cash flow, and sovereign, strategic and private-credit capital may keep flowing. Labs can also renegotiate commitments or raise equity, so stress could mean consolidation or a correction rather than collapse.
• Premise 3 (Medium impact) — Vendor financing and strategic equity stakes are common in infrastructure build-outs. Circularity matters only to the extent it is not backed by third-party revenue, and the argument does not separate the circular portion from the organic portion.
• Premise 2 (Medium impact) — Productivity gains from general-purpose technologies lag investment by years, and gains in coding and support are hard to see in aggregate data. The claim about Google's core business is contested and may reflect other forces such as ad cycles, regulation or search-market shifts.
• Conclusion (Medium impact) — Historical precedents (railways, fiber, cloud, early Amazon) show that heavy subsidy, circular financing and a later bust can coexist with durable industries. Fragility does not show that the underlying demand is illusory. The undated prediction also cannot distinguish the bearish story from an ordinary boom-and-bust cycle.

Suggested Improvements

• Premise 4: The share of revenue that is subsidized or circular is unmeasured, and subsidized pricing is conflated with demand that would vanish. — Break lab revenue down by customer type and funding source. Show retention and churn as prices rise, and use gross margin data to show whether the subsidy is temporary. This is the load-bearing link between financing strain and falling revenue.
• Assumption 3: The prediction has no date, threshold or trigger. — State a window and a measurable condition, such as a fall in lab revenue growth or a given drop in hyperscaler AI capex within a set number of years after funding tightens. A claim that cannot be wrong at any particular time carries little evidential weight.
• Premise 1: One expected rating is used as a proxy for the industry's financing, and the claim about OpenAI is asserted. — Give the actual rating, issuer and instrument, with spreads, lender participation and maturities. Separate labs and neoclouds from investment-grade hyperscalers. Lab credit and hyperscaler credit differ greatly, and an expected rating is not a final one.
• Premise 3: The share of hyperscaler and supplier growth tied to funded labs is not quantified. — Quantify circular flows from filings, using the share of cloud backlog and chip revenue tied to funded labs, and the share of lab funding from vendors. Circularity is only decisive if it covers most of the growth and is not backed by third-party revenue.
• Premise 2: The claims about payoff and the Google core business have no metrics or counterfactual. — Cite specific return-on-investment surveys, productivity data and search metrics compared with peers or a no-AI baseline. Lagging diffusion and measurement problems are an obvious alternative explanation.

Scenario Tests

• The rating arrives, but the labs keep raising large amounts of debt and equity at tolerable spreads while enterprise revenue keeps growing. (Challenges) — Financing would be dearer but functioning, which fits a leveraged growth story and weakens the claim that the resources will run out.
• Labs raise prices after funding tightens, and customer cohorts mostly stay. (Challenges) — Subsidy would prove transitional, and tighter credit would slow growth without causing revenue to fall.
• After a lab struggles to refinance, its suppliers and data-center partners face stress, and orders for compute fall sharply. (Supports) — This would show the tight coupling between lab solvency, supplier revenue and capex, and give real support to the spending-decline claim.
• A large share of lab usage turns out to come from startups spending venture money, and it drops quickly when their funding ends. (Supports) — This would show that demand depends on subsidy and would make the revenue-decline step plausible.
• A shakeout removes weaker labs and overbuilt capacity, but surviving leaders keep growing revenue. (Neutral) — Spending would fall in places, but the contraction of an industry whose revenue was mostly subsidy would not be demonstrated, and the business-or-not label would be undecided.

Coherence & Relevance

The argument is a coherent cumulative case for financial fragility, built from a financing strand, a payoff strand and a circular-demand strand. It becomes incomplete at the final step, which treats a funding problem as a demand collapse without a stated mechanism or threshold. The timing is also left open, which makes the claim hard to check.

• The expected junk rating means Anthropic will finance its commitments at higher cost from a smaller pool of lenders, with more risk on the suppliers that back it, so its compute commitments weigh more heavily as liabilities, and OpenAI faces the same problem in a worse form. (Moderate) — It bears on rising financing cost for the labs. It does not show that the cost will exceed what revenue or deep-pocketed backers can cover, and it is about one lab rather than the whole industry.
• So far the evidence shows AI paying off mainly for the companies that sell it, while at Google it is slowing the core business it was meant to strengthen. (Weak) — Weak payoff outside the sellers is not shown to cause buyers to stop paying, and it needs a premise that spending will stop when returns are absent. Adoption lags are not addressed.
• 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. (Moderate) — It shows concentration and circularity risk, but not the share of revenue that is independent. Circular revenue is not necessarily unreal demand.
• A large share of AI lab revenue rests on demand subsidized by investors rather than earned through returns, so it cannot be repeated in a normal economy; when venture money and cheap debt tighten, revenue will fall, which makes the industry an investment operation more than a business. (Strong) — It is directly on point, but its key quantities are missing: the subsidized share, the price response of customers, and the ability of labs to adapt. It also comes close to restating the conclusion.

View this argument on LogicFirst.ai