The Nvidia and OpenAI circular deal: how money loops inside the AI bubble

What did Nvidia and OpenAI announce in September 2025?

On 22 September 2025, NVIDIA announced a letter of intent with OpenAI for a strategic partnership. The announcement says OpenAI will build and deploy at least 10 gigawatts of AI data centres with NVIDIA systems. It says NVIDIA intends to invest up to $100 billion in OpenAI progressively as each gigawatt is deployed (NVIDIA newsroom, 22 September 2025).

The first gigawatt was planned for the second half of 2026 on the NVIDIA Vera Rubin platform, and the announcement describes the details as still to be finalised. This guide relies on that primary text alone. Terms can change between a letter of intent and final agreements, so read the latest company filings for the current position before you rely on any figure here.

How can money loop inside the Nvidia and OpenAI deal?

Take the announced ceilings and divide: up to $100 billion across at least 10 gigawatts averages up to $10 billion of investment per gigawatt. The investment arrives in stages, as each gigawatt is deployed, and each gigawatt deployed means NVIDIA systems in OpenAI data centres. The supplier’s capital and the supplier’s product therefore travel together.

A loop exists to the extent that the invested dollars pay for the NVIDIA systems. The announcement leaves that share open. If a large part of each stage returns to NVIDIA as payment for systems, NVIDIA’s revenue rises with its own investment. If OpenAI pays mostly from its own sales and outside funding, the loop is thin. The share is the number to look for.

How does the Nvidia and OpenAI deal differ from an ordinary chip sale?

In an ordinary sale the buyer pays from its own cash and the seller books revenue. Here the seller also becomes an investor, and the investment is staged against deployment. That link ties the supplier’s capital to the customer’s purchases, which makes the arrangement circular in structure and sets it apart from a plain order.

A letter of intent adds flexibility. Each side can adjust volumes, timing, and terms before final agreements, and each stage can be paced to results. Flexibility lowers the risk for both parties, and it also means the headline figures are ceilings. Treat the words up to and at least as the outer bounds of a range, and wait for filings to show the realised amounts.

How much of the Nvidia and OpenAI deal rests on outside demand?

Outside demand means customers who pay OpenAI for its products. The more of the data centre bill that such customers cover, the more the deal reflects real demand. Look at OpenAI’s reported revenue, its funding rounds from other investors, and its commitments to other suppliers, and compare them with the capacity it plans to deploy.

Then compare the timeline. Capacity arrives in gigawatts over several years, while customer revenue builds quarter by quarter. A wide gap between the two means the loop and outside funding carry the build-out in the meantime. The EQOS Asia AI study records about $200 billion a year invested across the sector (April 2026), so a single pairing with a ceiling of $100 billion is large next to that yearly total.

How does the Nvidia and OpenAI deal show up on each company’s books?

For the supplier, two lines move: revenue from systems and an equity stake in the customer. For the customer, two lines also move: new capacity and new capital. If the stake is valued at the customer’s latest funding round, the supplier’s stake rises when the customer raises money at a higher price, and that gain can sit beside the supplier’s sales.

That feedback is the valuation side of the loop. A higher valuation for the customer supports more funding, more funding supports more purchases, and more purchases support the supplier’s revenue and its own valuation. Each link is plausible alone, and together they can lift prices faster than outside demand does. Check how each company values and reports the stake.

Which filing items reveal how the Nvidia and OpenAI loop works?

Five items answer most questions: the definitive agreements, related-party disclosures, purchase obligations, customer concentration, and the valuation of equity stakes. Search each company’s quarterly filings for them and note the date and amount. The definitive agreements show whether the letter of intent became binding and on which terms.

Customer concentration is the second key. A supplier that reports a large share of revenue from a few customers faces more risk if one of them slows. A customer that relies on one supplier faces the same risk in reverse. Track the concentration figures each quarter, because a rising share tightens the loop and a falling share loosens it.

What should a decision maker take from the Nvidia and OpenAI deal?

Take the method, and leave the verdict on any company to the filings. Ask who invests, who buys, how much of the payment returns to the investor, and who else funds the buyer. The EQOS Asia method page shows how the market score of 0.337 is built, and loops raise the fragility behind it when each link depends on the next.

If your business sells to, buys from, or lends to AI firms in Singapore, ASEAN, or Hong Kong, a 30-minute pre-diagnostic with EQOS Asia scores the structure of your exposure. You can also test your inputs on the Λ simulator first.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile