Churn Chat (December 2025)
The Capital Cycle is the official podcast that Marathon Asset Management (the London firm founded in 1986) launched in 2024, hosted by financial historian Edward Chancellor, who interviews Marathon's investors about each Global Investment Review letter — applying the firm's long-term, contrarian "capital cycle" supply-side approach.
In plain words
This piece uses a fictional board meeting to reveal a dangerous game in AI infrastructure: companies commit trillions (like $1.4 trillion) to build computing power, but the money comes not from paying customers but from circular stock deals, paying chipmakers to buy their own chips, and issuing AI-rated bonds (bonds rated by AI itself). For ordinary investors, if an AI stock's revenue relies on storytelling and capital recycling rather than real customer cash, it's likely a bubble waiting to burst. Worth reading because it helps you spot companies that look shiny but are really just passing the parcel.
This report, presented in the form of a fictional PonzAI board meeting, critically exposes the "Ponzi scheme" characteristics of AI infrastructure investment. CEO Stanley Churn acknowledges that the company has committed approximately $1.4 trillion to build 26 GW of capacity (equivalent to 20 nuclea
Theme and Background
This chapter uses a fictional board meeting at PonzAI to reveal the pervasive "Ponzi financing" characteristics in the AI infrastructure investment space—where investment decisions are based on expectations of continued future funding rather than genuine business cash flow. The report argues that the current market environment is enabling this kind of "pass-the-parcel" capital game.
Core Thesis
The author's central judgment is that the AI infrastructure investment frenzy is essentially a capital fantasy game, where participants rely on constant storytelling, equity issuance, vendor financing, and cross-shareholding among peers to sustain their cash flows, rather than on real customer revenue. Contrary to the consensus, the author does not view AI investment as a "future productivity revolution" but characterizes it as financial engineering lacking cash flow support.
Key Arguments and Data
- Massive capital commitments: PonzAI has committed approximately $1.4 trillion over the next 8 years to build 26 GW of computing capacity—equivalent to 20 new nuclear reactors or the entire GDP of Spain.
- Lack of genuine cash flow: The CFO explicitly admits that the primary source of funding is "circulate" (Circulate)—buying chips to boost the stock price of partner GameChip, then having SoftHead sell chip stocks to buy PonzAI shares, creating a closed loop.
- Vendor financing (VendiFi): Chip companies actually pay PonzAI to buy their chips—i.e., "paying someone to buy your own product."
- Warrant plan (Levitate): PonzAI receives warrants in the chip companies it invests in when purchasing chips.
- Equity in exchange for usage rights: The right to use ChurnGPT 6 is exchanged for equity in Scone Holdings, with no cash inflow.
- IPO strategy: Plans a $1 trillion IPO, projecting 3 billion users by 2035, 50% global productivity improvement, and 40% EBITDA margin. The CTO jokes, "Aren't we a non-profit?" The CEO replies, "That saintly label is just PR."
- Backup plan: Issuing AI-rated bonds (DUMBOs, AI-certified bonds), even using the government as the ultimate backstop via "public-private infrastructure investment."
| Financing Tool / Strategy | Specific Content | Nature |
|---|---|---|
| Circulate | Buy chips → boost chip stock price → SoftHead sells chip stocks to buy PonzAI shares → uses money to buy more chips | Stock-for-stock, no external cash |
| Vendor Financing (VendiFi) | Chip companies pay PonzAI to buy chips | Reverse cash flow, vendors bear the cost |
| Warrant Strategy (Levitate) | Obtain warrants from chip companies when purchasing chips | Equity swap, non-cash income |
| Equity for Usage Rights | Exchange ChurnGPT 6 usage rights for Scone Holdings equity | Service offsetting equity, no revenue |
| DUMBOs Bonds | AI-certified bonds issued by AI rating agencies | AI rating AI, credit self-loop |
| $1 Trillion IPO Plan | Assumes 3 billion users, 50% productivity improvement, 40% EBITDA by 2035 | Extremely optimistic linear extrapolation |
| Government Backstop | Secures tax support under the guise of "national infrastructure" | Shifting risk to taxpayers |
Companies/Assets Involved
- PonzAI (fictional): Core subject; the author is clearly bearish. The company promises massive investments but has no genuine cash flow, fully dependent on capital games to operate. The CEO bluntly states, "If ChurnGPT 6 fails, we'll pitch the next-generation product."
- GameChip (fictional): A chip partner; the author suggests its stock price is artificially inflated by PonzAI's purchasing activity, indicating a false market cap bubble.
- SoftHead (fictional): The key "relay runner" in the circulate trade, selling chip stocks to buy PonzAI, acting as a capital intermediary.
- Scone Holdings (fictional): The investee in the equity-for-usage model, also with no cash income.
Investment Insights
1. Avoid companies that rely on future cash flow promises. If an AI infrastructure company's revenue model heavily depends on equity circulation, vendor financing, or government backstops, rather than actual customer payments, its valuation is likely wildly overstated.
2. Beware of "flywheel models." If the sole driver of stock price appreciation is the company spending its own money to buy supplier stocks—not margin expansion or user growth—the stock carries systemic crash risk.
3. Focus on genuine sources of cash flow. Investors should strictly differentiate instruments like "DUMBOs" bonds from traditional asset-backed securities. AI-rated credit products lack independent, trustworthy cash flow verification and represent risk self-looping.
4. Be wary of overly optimistic projections (e.g., 40% EBITDA, 3 billion users). Such forecasts are typically used as IPO marketing tools, not based on verifiable orders or retention data.