Imagine you make industrial ovens. A startup bakery wants twenty of them and can only afford five. So you buy 10% of the bakery for cash, and that cash lets the bakery buy all twenty.

Two things now happen in two different places. The twenty oven sales land in revenue, where the market celebrates them. The cash you put into the bakery lands on the balance sheet as an investment, not on the income statement as an expense. The same dollar left your building, came back through the front door, and was recorded once as profit and once as an asset. This is not illegal, but your revenue growth and your investment outflows are the same event described twice in different languages.

That is the shape of the accusation against NVIDIA. The company has taken equity stakes in the customers buying its systems: the AI labs, and the "neoclouds" like CoreWeave and Nebius that rent GPUs to everyone else. With CoreWeave it went even further; NVIDIA contractually agreed to buy any compute capacity CoreWeave cannot sell to anyone else, through 2032. That’s a promise to become the customer of last resort for your own hardware if real demand falls short.

Vendor financing is ancient and mostly legitimate. Boeing helps airlines finance aircraft. Equipment makers have always lent to buyers, because the buyer's constraint is capital, not appetite. Seeding an ecosystem so your platform becomes the default is ordinary strategy, and here it has a commercial edge: a customer financed onto CUDA is a customer not buying from AMD or designing custom silicon with Broadcom.

Scale matters too. NVIDIA has generated close to $200 billion of cash in two years against tens of billions deployed into partners, and the majority of its revenue still comes from Microsoft, Amazon, Google and Meta writing cheques out of operating cash flow with no NVIDIA money anywhere near the transaction.

So "NVIDIA is selling to itself" might be too strong of a statement. The more accurate thing to say is that circularity is a dial, not a switch, and what it tells you about is the quality of revenue. A dollar supplied by NVIDIA is a lower-quality dollar than one from a customer's own cash flow. It is more cyclical, more reflexive, and it carries a hidden liability: if the buyer fails, NVIDIA loses the equity, loses the future revenue, and in CoreWeave's case has to write the cheque for the unsold capacity as well.

Which is what makes last week's announcement so significant. NVIDIA signed memorandums of understanding (MOU) with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to build "independent compute financing platforms" intended to mobilise over $500 billion of third-party capital, creating dedicated pools of money at attractive rates for NVIDIA's customers. The company is adamant that this is the answer to circularity, not an instance of it. Outside institutions underwrite each project themselves, while NVIDIA supplies the platform rather than the capital.

None of it is committed money yet. These are MOUs, and the decisive terms (rates, maturities, how much support NVIDIA itself ends up providing…) appear only in final agreements.

The deeper question is whether GPUs deserve to be treated as an asset class at all. Jensen Huang's pitch is that a GPU cluster should be financeable the way a toll road is: broadly adopted, transferable between operators, and kept productive for longer by continuous software improvements. If lenders accept that, an AI datacenter can be funded against its own contracted cash flows rather than its owner's balance sheet. This changes everything, because the number of companies able to fund tens of billions of capex from operations is about five. Project finance opens the market to everyone else.

But a toll road does not get obsoleted by a new generation of road every eighteen months. That CUDA keeps older silicon earning is the load-bearing assumption of the entire edifice. If it holds, this is normal infrastructure credit with unusual collateral, and the AI buildout has just found a far deeper pool of funding. If it doesn't, a group of pension funds and insurers will write a huge amount of loans against a rapidly depreciating asset… in a market where the depreciation schedule is set by the borrower's own supplier. In other words, a recipe for disaster.

Upgrade to unlock full access to the RatedA Portfolio (with weekly updates on every buy/sell and rationale), and the complete database of companies + ratings.