Nvidia Is Now on Four Sides of Its Own Demand

Across two days at the end of July, Nvidia appeared in our deal tracker four separate times. In none of the four was the role "sold some GPUs."

On July 27 it was putting roughly $5 billion of equity into Safe Superintelligence, with access to its next generation Vera Rubin systems attached to the deal. The same day it was negotiating a $250 billion financing guarantee to back OpenAI’s lease and construction debt on a 10 gigawatt data center. A day earlier it had committed to a partnership with SK Group valued above $500 billion to build a 2 gigawatt AI factory in Korea and co-develop next generation HBM memory. And it was convening more than 30 technology and security companies into the Open Secure AI Alliance to write shared standards for open models.

Investor. Guarantor. Landlord. Standards body. All in one week, from the company whose chips every one of those buildouts is designed to hold.

## The part that should make you slow down

There is a name for a supplier financing its customers’ ability to buy: vendor financing. It is not inherently improper, and at Nvidia’s current margins it is affordable. But it does one specific thing to the numbers. It makes demand less independent of the seller.

A backlog built partly on capital the vendor supplied is not the same signal as a backlog built on customers spending their own money. The telecom equipment makers learned that at the end of the 1990s, when the loans they wrote to buy their own gear came back as writedowns. Nobody is claiming the analogy is tight. AI compute has real, observable utilisation behind it in a way that a lot of unlit fibre never did. The point is narrower: when the same balance sheet sits on both sides of a transaction, the demand figure stops being a clean read on the market and becomes partly a read on the vendor’s risk appetite.

## What our own counts say about scale

Some perspective from the tracker. Of the 416 AI deals we have recorded so far in 2026, compute agreements are 58 of them, about one in seven. Straight investments are 201, and partnerships 91. Compute is a minority of AI dealmaking by count while carrying almost all of the headline numbers, which is worth remembering the next time a gigawatt figure gets treated as the state of the whole industry.

The counterweight is visible in the same data. CXMT’s Shanghai listing raised at least $8.6 billion and rose over 450 percent on debut. That is a memory manufacturer funded by public markets rather than by its chief supplier. Two very different ways of financing the same buildout, running at the same time.

## What to do with this

**Read the filings, not the announcement.** The interesting disclosure is whether guarantees like the OpenAI one sit on Nvidia’s balance sheet or beside it, and how equity stakes in customers are carried. That detail decides how much of the reported demand is genuinely arms length. It will show up in a filing long after the press release stops trending.

**Ask one question of every large compute announcement: who is paying, and is the payer also the seller?** It is a fast filter, and it sorts the market into two piles that behave very differently when funding tightens. You can check the counterparties on any deal we track in the deal tracker.