The wire left Redmond on January 23, 2023. $10 billion, headed for a nonprofit called OpenAI. Two years later, Microsoft's total commitment had reportedly climbed past $13 billion, and OpenAI had signed procurement deals measured in tens of billions of dollars. The capital traveled one direction. Then it came back dressed as revenue.

That recycling sits at the heart of a new analysis from The Conversation titled "Who's paying whom? The circular money machine behind the AI boom (or bubble)." The piece picks apart the closed loops now standard across AI's capital stack. Its central question is bracingly simple: when vendors invest in their own customers, who is actually paying for the buildout?

How the loops are built

A textbook case runs like this. A chipmaker writes a check to a model lab. The lab signs a multi-year purchase agreement for the chipmaker's accelerators. Each side books the deal as growth. Each side also books the investment as a strategic stake. The numbers, on paper, double.

The pattern repeats across the stack. Hyperscalers fund AI startups through cloud credits and equity rounds. Those startups then burn their runway on the same hyperscalers' compute. The same dollars, dressed in different accounting lines, count twice. Oracle's expanded Stargate commitments and the AMD–OpenAI agreement, both disclosed in 2025, fit the template.

Why the timing matters

The Conversation's framing, with its parenthetical "(or bubble)," mirrors a debate that has crept from industry Twitter into regulator offices. The Federal Trade Commission's ongoing review of the Microsoft–OpenAI arrangement, and the European Commission's parallel scrutiny, reflect a new willingness to ask whether vendor financing has blurred the line between supplier and customer.

The IMF, in its October 2024 World Economic Outlook, flagged AI-related valuation concerns. The Bank of England echoed similar warnings. Sell-side analysts at major banks have used the word "circular" in investor notes over the past twelve months.

What "circular" really risks

The concern is not fraud. Each deal, taken alone, is real. The concern is opacity. When revenue depends on capital that originated on your own balance sheet, the underlying economics get harder to read. One tightening of the loop, and the chain can rattle. Vendor financing, once a niche corner of chip finance pioneered by ASML and TSMC, has scaled into the dominant funding model of the AI buildout.

The Conversation's contribution is naming the structure. Whether that structure is scaffolding or a house of cards is left, sensibly, to the math — and to the regulators now circling it.

What comes next

The next stress test arrives with earnings. The four largest U.S. hyperscalers have guided combined capital expenditures above $200 billion for 2025. If that spending holds through 2026, the circular thesis fades into background noise. If it falters, the loops become the story. Watch the cloud revenue lines, not the press releases. The numbers will tell.