The Risk Now Has a Sales Channel. Should We Worry?

The number sold to you is five hundred. On August 10 Nvidia announced memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize more than $500 billion of third-party capital to fund the machine. No fund exists. None of the six has said what it is putting in, there is no timetable, and the release closes on the line no headline picked up: the partnerships remain subject to the execution of final agreements.

The number that counts is one hundred and twenty-five. Huang has said Nvidia will have the option to backstop up to $125 billion, 25 percent of the potential deals. That signature is what makes the paper sellable.

The word to hold on to is the one David Solomon chose in the release: distribution. Goldman Sachs says it is excited by the new opportunity to create a market for credit backed by Nvidia compute. A market has two sides. On one is the circle that needs to shed exposure. On the other are you, usually without knowing it, carrying big tech's bet on your own back.

Who the six are explains the rest. Apollo manages roughly $1.05 trillion and owns Athene, which sells annuities and retirement products. BlackRock is the largest gatherer of pension savings on the planet. Morgan Stanley puts global data-center capex at $2.9 trillion between 2025 and 2028, with a financing gap near $1.5 trillion, about $800 billion of it expected from private credit. Private credit presents itself as an alternative asset class; in practice a great deal of it is insurance balance sheet. Policy reserves. Money that has to pay someone an annuity in twenty years.

The proportions help. In the year ended January 25, 2026, Nvidia booked $215.9 billion in revenue, up 65 percent, with $193.7 billion of it from data centers. The $125 billion in guarantees is worth more than half a year of sales. Add the roughly $250 billion under discussion for the OpenAI campus in southern Ohio and the potential exposure runs past the entire annual turnover of the most valuable company in the world. That is the point where a structure grown too circular stops being a private problem.

An engineer looking at a structure like this hunts for the place where the timelines fail to line up. Here it is the collateral. The release promises long-duration, usage-linked revenue, which means long loans. The security underneath is GPUs, and on the useful life of GPUs the industry disagrees with itself: Microsoft stretched its depreciation schedule from four years to six, Meta sits at five and a half, Nvidia says customers use four-to-six-year lives, while Michael Burry argues the real economic life is two or three. Meanwhile the GPU-backed loans you actually see in the market run two to four years with 20 to 30 percent annual write-downs, which tells you how the people putting up the cash have done the math. A long loan resting on an asset that ages fast is a textbook duration mismatch. It is the piece that gives way first, and it pays to know where it sits before it does, because the GPUs they are talking about have not been invented yet.

Then there is the thing this architecture does that no chart shows. Moving the exposure off Nvidia's balance sheet and into independent vehicles changes where the risk appears and leaves its size untouched. It is the move of someone who fixes an error by taking it out of the log: the component keeps failing, it simply stops showing up in the monitoring anyone reads. Nvidia sent its analysts a memo denying that any of this is vendor financing. To the letter that is true, and that is what makes it interesting: the guarantee stays a contingent liability, and the financed part migrates into structures whose total no annual report will ever give you.

It should be said that financing heavy infrastructure with third-party capital is a serious and old trade. Railways, power grids, undersea cables: none of them were built out of the cash flow of the firms selling the equipment. Turning compute capacity into a financeable asset class is a legitimate idea, and the six who signed know their business. One condition holds: that final demand arrives at the end of the chain, from people paying to use the product rather than to build it. Infrastructure repays itself through tolls. The enthusiasm of the company selling the asphalt appears nowhere in the repayment schedule.

This is where the story stops being about Wall Street alone. Until yesterday the risk left the circle by leakage, through index funds, taxpayers, the people paying the grid bill. Now there is a pipe with names on it, and pensions sit at the far end. Anyone working inside this wave would do well to read every billion-dollar announcement the way you read a runbook: which component fails first, who notices, how long before anyone gets paged. When you are told AI demand is insatiable, look at who signed the order and who is holding the note. They are rarely the same person, and one of them is you.

For two years the question was whether the circle would hold. Now the circle has a teller window, and it is open to the public.