Circular Investing in AI Infrastructure

In the AI Story

The pattern has been identified by multiple analysts, including Paul Kedrosky (who called it the AI infrastructure Minsky moment), Howard Marks of Oaktree Capital (in his 2025 memo "Is It a Bubble?"), and the Certuity research group (in their October 2025 analysis "Are We in an AI Bubble?"). The analyses converge on a common observation: the revenue streams that justify AI infrastructure investment are substantially generated by the same ecosystem of companies making the investment, creating a self-reinforcing loop whose stability depends on continued external validation.

The specific structural features identified include vendor financing loops in which chip manufacturers finance the customers purchasing chips; compute-credit arrangements in which cloud providers extend credit to AI companies that will return the capital as compute spending; and capital expenditure projections — the estimated 1.3 percent of U.S. GDP in hyperscaler AI capex in 2025, rising to 1.6 percent in 2026 — that would require substantial external enterprise adoption to justify. The MIT Media Lab's finding that ninety-five percent of enterprises reported zero measurable AI ROI through August 2025 suggests that the external adoption has not materialized at the rate the investment assumes.

OpenAI's financial position illustrates the pattern at its most extreme. The company has committed to spending $1.4 trillion over eight years on data center infrastructure, against $13 billion in annual revenue. The company projects $74 billion in operating losses in 2028 alone. The cash flows cover neither interest nor principal on the commitments. The position is sustained entirely by the market's belief that AI's value will continue to appreciate at a rate that eventually justifies the investment — which is, in Minsky's precise technical sense, a Ponzi position.

The circular pattern is not unique to AI. Kindleberger documented analogous structures in the South Sea Company, Japanese keiretsu, and the dot-com era's business-to-business revenue recognition. The pattern is a recognized feature of mature speculative phases, and its presence is one of the empirical signals that a boom has moved from genuine displacement into euphoric territory.

Origin

The specific identification of circular investing in AI infrastructure emerged in the 2024-2025 financial commentary cycle, with Paul Kedrosky, Howard Marks, and various equity analysts converging on the framing. The pattern's mapping to Minsky's Ponzi category was developed in Marks's investment memos and in the Certuity research group's public analyses.

Key Ideas

Endogenous revenue. The revenue justifying investment is substantially produced by the investment itself — the signal and the noise are generated by the same mechanism.

Vendor financing. Chip manufacturers extending credit to customers, cloud providers extending compute to AI companies — structural features that Kindleberger identified in previous manias.

Capex-revenue divergence. Trillion-dollar infrastructure commitments against billions in current revenue, sustained by projections of future appreciation.

External validation gap. MIT Media Lab data showing ninety-five percent of enterprises reporting zero measurable AI ROI — the external validation required to transform speculative positions into hedge ones has not materialized.

Minsky Ponzi classification. The infrastructure positions meet the precise technical definition of Ponzi finance: income covers neither interest nor principal, and the position is sustained by expected asset appreciation.

Debates & Critiques

Defenders of the current investment pattern argue that the infrastructure buildout will generate returns as AI capabilities mature and enterprise adoption accelerates. Critics, following Minsky, argue that the pattern's structural features — circular revenue, vendor financing, capex-revenue divergence — are the empirical signatures of speculative phases that historically resolved through significant corrections. The debate is unresolved; the data through mid-2026 is consistent with both interpretations, which is itself a Minskyan observation about the invisibility of fragility from inside the boom.