Sunil Venkataram
Back to garden
•
🌰 seedling
Circular Financing as a Bubble Signal in AI
type/noteinvestingaieconomics

Circular Financing as a Bubble Signal in AI

The pattern

The template is simple:

  1. Company A (the “investor”) puts capital into Company B (the “counterparty”).
  2. Company B agrees to buy a significant amount of goods or services from Company A.
  3. Company A books the purchase as revenue.
  4. The net cash movement from the outside world is much smaller than the gross revenue figure suggests.

The arrangement can be legitimate when there’s a genuine strategic rationale, like a platform investor supporting an ecosystem partner. It becomes a warning signal when the revenue it generates is a material portion of the “investor” company’s top-line growth, when the same pattern repeats across multiple counterparties at the top of an industry, or when the accounting treatment makes the arrangement look cleaner than the underlying cash flows.

The specific examples

From the public record as of 2026:

  • Microsoft ↔ OpenAI. Microsoft invested ~$13B+ in OpenAI; OpenAI has committed to very large purchases of Azure compute. The net cash flow from OpenAI to Microsoft over the arrangement’s lifetime is some fraction of Microsoft’s announced OpenAI-related Azure revenue.
  • Nvidia ↔ AI startups. Nvidia has invested in multiple frontier AI companies. Those companies commit to purchasing future Nvidia compute capacity. The same dollar of Nvidia equity capital flows out as investment and flows back as compute purchases.
  • Amazon ↔ AI startups. Similar arrangements reported with multiple AI companies and AWS compute.

Bill Gurley characterizes this on the Tim Ferriss podcast as “precisely questionable”, stopping short of calling it fraud but noting that clean accounting would present these arrangements differently.

Why this is a bubble signal, not just weird accounting

Three reasons circular financing is a late-stage bubble indicator:

1. It inflates reported growth beyond what end-customer demand can sustain. When Nvidia books revenue from an AI startup using Nvidia’s own invested capital to pay for the compute, that revenue signals balance-sheet recycling, not independent demand. End-user demand may also exist, but the revenue figure muddles the two.

2. It creates correlated downside when the cycle breaks. If the AI startup can’t raise its next round or pivots away from heavy compute use, Nvidia loses both the revenue and the invested capital in the same event. The conventional assumption that equity investments and supply-chain revenue are uncorrelated breaks down exactly when it matters most.

3. It precedes previous industrial-bubble corrections. In the late-1990s telecom bubble, equipment makers (Lucent, Nortel) financed their customers’ equipment purchases. When the customers couldn’t pay, the vendors took write-downs that exceeded any prior guidance. The pattern repeated with fiber-optic capacity: telcos bought each other’s bandwidth through circular IRU (indefeasible right of use) swaps that created paper revenue without net economic activity. Both ended in multi-year corrections that erased most of the peak market cap.

The indicator to watch

The telltale in every historical circular-financing episode is the same: revenue quality deteriorates before revenue quantity does. Investor presentations start emphasizing “committed revenue” or “contracted revenue” rather than actual cash collected. Accounts receivable grow faster than revenue. Net cash from operations lags reported earnings.

In the current AI cycle, these signals have started appearing in specific filings but haven’t yet triggered broad market repricing. The question is when the market starts pricing in the quality discount.

What to do about it

  • For equity analysts: build a “net of circular financing” revenue adjustment for every AI supply-chain company. The adjustment should subtract revenue booked from customers the company has itself invested in. This gives a cleaner picture of end-user demand.
  • For investors: weight companies whose revenue is primarily from non-portfolio customers higher than companies with significant circular exposure. The former are more likely to survive a correction intact.
  • For founders: be cautious about strategic investors whose check size implies compute commitments you cannot fundamentally justify. The conflict of interest is real, and reputational damage from appearing in a circular-financing story compounds if the cycle breaks.
  • For regulators: historical precedent suggests this is an area where disclosure requirements will eventually tighten. Companies that get ahead of that with clean disclosure will be rewarded by the market during the correction.

The honest caveats

  • Some circular deals are legitimate. A platform investing in an ecosystem partner can reflect genuine strategic alignment. The judgment call is whether the structure inflates reported numbers or reflects a real business relationship.
  • Circular financing is an indicator and accelerant, not a root cause. The root cause is always demand reality catching up with speculative pricing.
  • Some circular financing will clear cleanly. If OpenAI’s end-user demand materializes as projected, the Microsoft-OpenAI arrangement will look like a smart platform investment rather than a circular revenue scheme. The retroactive judgment depends on whether the downstream economics actually show up.
Backlinks (3)
Connected Notes