Circular Financing, and One Company in the Middle of It

Chapters

• 0:00 The frame
• 1:14 What does this company do?
• 1:50 What does a shareholder own?
• 2:48 Can the machines pay for themselves?
• 4:47 Is this circular financing?
• 6:43 Is it a bubble? Will they fail?
• 7:42 What I do not know

The frame

I read filings to understand companies and see how they work.

What we’ve been trying to learn this year is circular financing. What it is and what is good and what is bad about it.

CoreWeave sits in the middle of that, in the middle of the whole AI build-out argument, which makes it an interesting read. This week we came at it from the supply chain, the company’s role next to other players, and what we found surprised us.

From that angle, this looks like a financing source for the supply chain more than an operating company.

Here’s Jensen Huang on the Goldman stage earlier this month on how these deals get funded.

The offtake’s job, in that sentence, is to make the financing come together.

One thing before we start: nothing here is an accusation. Everything I’m going to show you is disclosed in public, and where the company cannot answer something, it says so. What I’m interested in is the part the disclosure does not reach.

So let’s start at the beginning.

What does this company do?

Most companies borrow to buy something that earns. In the first half of this year, this one raised about $20 billion and billed under five.

Four dollars in the door from lenders and investors for every dollar it sold.

And what it raised, net, went almost exactly into the machines. So most of the money moving through this company is money being raised and repaid rather than money being earned.

And that’s what the filings say once you read them for cash rather than for profit.

What does a shareholder own?

They own the machines after the loans are paid off, and the loans work like a car loan. Monthly payments, six years of them in this case, and at the end the principal of the loan is zero, and you own the thing free and clear.

But it’s a used machine, six years of work on it, the way a car gets miles put on it. And if it was well built and well cared for, and let us assume both here, it may have miles to give.

That is what a shareholder owns: a six-year-old machine with some life in it after everybody else has been paid.

And just like a car, what matters is what it will fetch then. The difference is that for cars there’s a book. Somebody publishes the number. For these, there is no book.

Every market built to price computing stops inside three years. The question that settles this company sits out six years.

Can the machines pay for themselves?

For those six years, the machines have to be fed. Power, like gas in a car, maintenance, and somewhere to keep them.

All that comes out of what they earn, and there is a real question in this case about what those net earnings are going to be.

The bank doesn’t care. It wants its monthly payment on time. If the machines do not earn their keep, the company still makes the payment out of cash. So how much is left over?

The company files that answer. The machines in service throw off cash equal to about 8.968% of the money sitting in them. The newest money the company borrowed costs 9.12%.

It doesn’t seem like a very large gap. It’s about $79 million, actually, and it only covers the interest. Because, as we know, a car payment is interest and a piece of the principal. That’s why at the end of the loan the principal is zero, because you paid it back during the process.

And principal appears nowhere on a profit statement. Add it back and the payments run $2 to $5 billion a year more than the machines bring in.

There is something on the profit statement that gets at this, and it’s called depreciation. And it’s an estimate of the wear and tear on the machine. It’s an estimate. Most companies ask you to look past it. It’s a number that they call adjusted EBITDA.

And here the estimate is a live question, because nobody knows how long these machines last, or how they will work in six years, or what they will be worth then.

It’s all done quite proper. It’s quite completely disclosed. It’s just that the question an owner would most want answered, which is whether it was worth it at the end of six years, is not answered anywhere. And the company is candid about that, too.

Is this circular financing?

It’s all circular. A bank lends you money for a house and then owns paper backed by the house. A supplier finances the customer who buys its product.

It’s ordinary and it’s most of finance.

Think about why and start at the bottom. Cash on the barrel is the one trade that’s not a loop. The coin goes on the counter, the good comes back. It’s finished, it’s complete.

The money is just a coin. It has no character on its own. It turns into something the moment you give it a character. Count it, store it, promise it, use it for something. Now it’s a thing in its own right. It can do its own thing, and it can come back around.

An IOU does that, a trade payable does that. A bank doesn’t necessarily have to be involved for something to be financed.

So after that, it’s only a matter of complexity. How far you are from the coin on the barrel.

So one fairly straightforward example. A private company gets set in a funding round that’s funded by the people who funded the earlier rounds. That valuation then is used to get a loan by one of those founders or funders.

Every step is lawful, it’s totally disclosed, it’s completely normal, and the mark supporting the borrowing was set by the borrower’s own counterparties or partners.

That’s what to look at. Not whether something’s circular. It’s whether anyone outside the circle has priced it.

Is it a bubble? Will they fail?

Let’s start with what a bubble is. That’s a whole set of companies and industry collectively over-invested and overvalued against the whole of their opportunity. It’s a claim about an economy or a sub-economy. Filings don’t reach that far.

Crystal balls might, but findings don’t. What a filing supports is much narrower, and it is one company at a time.

So here’s the narrow version. There’s a cash schedule, and it’s written down, and the business as it currently runs does not cover it. In this case, between $2 and $5 billion a year on dates already in the documents.

Whether it gets met depends on what the machines earn and what the market will lend. Both are knowable later, and neither is knowable now. What I can do is give the number and say where and when to check it.

What I do not know

I’ll give you two examples here.

First, I don’t know what those machines will be worth in six years or what they will earn between now and then. That’s crystal ball work, and mine is no better than anybody else’s. It’s also the question the whole thing turns on, which is why I keep pointing at it instead of answering it.

Second, more immediately, I have a question about this company’s convertible note that’s due in 2031. It’s described three different ways in three different places. That’s completely normal. It’s almost certainly my retrieval and my reading rather than the filing, but until I have a clean reading, a clean filing collected together, those numbers stay out of anything I publish.