Who’s paying whom? The circular money machine behind the AI boom (or bubble)

Imagine a company that sells you equipment, invests money in your business and then rents back what you build with it. It’s your supplier, your investor and your customer all at once.

That’s known as circular financing.

Many artificial intelligence (AI) companies, including Microsoft, Amazon, Google and Nvidia, have adopted a version of this opaque method of financing.

A chip supplier invests in a customer such as a data center builder, the customer spends the money on the supplier’s chips, and both companies report growth.

One central bank analysis found that 55.2% of investments in AI firms came from other AI firms.

What’s not clear to investors or the market is how much of this AI boom is being driven by real, end-user demand—and how much might be a bubble.

It all comes back to Nvidia

Take Nvidia. The world’s leading chipmaker is investing in ChatGPT owner OpenAI to support data centers filled with Nvidia chips.

Nvidia will invest US$30 billion in a private funding round for OpenAI and provide a guarantee of up to US$105 billion for an OpenAI data center in Ohio. It will also invest in the company building it.

That helps OpenAI remain Nvidia’s best customer, but it also raises the question: Is the chipmaker propping up the market?

Nvidia has invested in other AI firms as well, including up to US$10 billion in Anthropic, the developer of Claude.

It bought US$2 billion in shares in cloud provider CoreWeave to help it speed up the build-out of AI factories that use Nvidia technology.

In total, Nvidia made more than 50 different investment deals in AI companies in 2024 alone.

Some of those deals are staged over years or still open-ended. But these circular deals help the AI build-out move faster than it could without Nvidia’s financing.

The Australian story—Firmus

Data center builder Firmus—which last week abandoned plans to list on the Australian Securities Exchange—provides the local example. Nvidia wears all three hats for Firmus:

1. Investor: Nvidia first invested in Firmus Technologies through the company’s A$330 million capital raise in September 2025 to finance data centers in Tasmania. It then participated as an existing investor in two further funding rounds, one in April 2026 and Firmus’ US$2 billion funding round in August 2026. Nvidia did not disclose how much it invested.
2. Supplier: Firmus plans to build data centers using Nvidia’s DSX AI factory platform, which combines chips, networking and software to provide computing capacity.
3. Customer: Firmus signed Nvidia as a customer for its data center in Batam, Indonesia, in a deal worth up to US$30 billion over six years. Meanwhile, OpenAI (with its existing relationship with Nvidia) has been named as an anchor customer at Firmus’ Malaysian data centers.

These circular deals also concentrate risk

The same loop concentrates risk. With Nvidia acting as investor, supplier and financier across the AI sector, trouble at one firm can spread quickly to the others. It also complicates valuations.

For instance, OpenAI’s method of reporting revenue has been questioned. Its shares were hit last week when the Financial Times reported its annualized revenue was US$20 billion less than the company had suggested.

The Reserve Bank of Australia warned in its recent Financial Stability Review that there are “pockets” of higher-risk companies, including those using opaque or circular funding arrangements, where chipmakers provide financing to the companies that purchase their products.

Warning about the possibility of the AI bubble bursting, the RBA said, “One possible trigger could be a shift in sentiment toward the AI investment boom, which is increasingly fueled by expectations of sustained rapid earnings growth and a debt-financing cycle that is becoming more opaque and circular.”

The concern is that nearly all the gains in the U.S. stock market have been driven by AI companies.

On Wall Street, chipmakers have contributed around 70% of the total gains in 2026 for the S&P 500, and Nvidia is now the index’s largest single company at about 8.06% of the index.

This makes the market fragile. If investors begin doubting that OpenAI and similar customers can earn enough revenue from customers to meet their commitments, Nvidia would be hit on two fronts: first, via lower expected sales, and second, via its financing exposure to those same companies. That would drag down the whole market due to Nvidia’s heavy weighting in the key index.

Australians would feel it too. Nearly 90% of households are exposed to AI stocks due to the U.S. shares held through their retirement funds.

A question of timing

None of this means the AI boom is fake. The issue is timing.

Nvidia, OpenAI and Firmus are betting that future revenue from users and businesses will grow fast enough to pay for today’s massive data center build-out. Until then, investors, lenders and the companies’ investments in one another fill the financing gap.

There is a historical comparison with the 1990s telecom boom. In what was then called the “New Economy,” equipment makers such as Cisco Systems financed customers to buy their equipment. Too much fiber was laid too soon, and investors lost heavily when the stock market bubble burst.

But that fiber lasted decades. The newest AI chips have a shelf life of only a few years. If the expected revenue and demand from businesses and consumers arrive late or fall below forecasts, shareholders and retirement funds will bear the losses.

Who’s behind this story?

Swati Mestri

Swati Mestri holds a bachelor’s degree in Electronics Engineering and has worked as a content editor since 2019. She has experience editing research documents across technology, health care, and materials science, and has a particular interest in technology and space.

Full profile →

Andrew Zinin

Master’s in physics with research experience. Long-time science news enthusiast. Plays key role in Science X’s editorial success.

Full profile →

This article is republished from The Conversation under a Creative Commons license. Read the original article.