Boosting revenue with vendor financing

AI companies made headlines recently because of circular vendor financing. Basically, Nvidia invests in OpenAI which buys services from Oracle which buys chips from Nvidia. Surprisingly, this setup lets you boost ecosystem revenue by over 100% of money spent. Let's look at how the accounting works.

What's going on

I think this Bloomberg article (archive link) from earlier this month spawned most of the discussion. It takes stock of the various deals inked recently and points out that there's an unusually large amount of A gives money to B, just for B to then buy stuff from A. The article has this nice graph that's very circular:

We're not going to be talking much about the specific deals made and instead focus on the how the accounting here works. If you're interested in the deals, the Bloomberg article linked above is good. Patrick Boyle has a video giving the rundown as well. For differing interpretations, here's an article in the American Prospect calling it a bubble and here's one in Owenomics explaining why it's not.

If you already know how a company's financial statements work and fit together like a puzzle, then feel free to skip to Boosting revenue.

Investments don't lower earnings

When looking at companies as investors, we usually talk about things like earnings and earnings growth. In the case of young companies without earnings, we usually focus on revenue instead. We think of companies as money making machines, so the most important aspects are how much money is going into them as revenue and how much is coming out as earnings or profit.

In 2024, Nvidia was a really good money making machine. Here's their income statement (from the 2025 Annual Report):

The way an income statement works is that it starts with Revenue as the top line, subtracts costs, and concludes with Net Income/Earnings/Profit as the bottom line. For the year ending January 2025, Nvidia had $130 billion in revenue and $73 billion in earnings.

Let's pretend for a moment that Nvidia's recent $100 billion investment into OpenAI happened last year. How would that impact Nvidia's earnings? Would they have have ended the year with a $27 billion loss?

The answer is no. Investments don't change earnings. In fact, investments don't appear on the income statement at all because the income statement is mostly concerned with a company's operating business.

In Nvidia's case, the income statement says:

• How much money they made from selling chips ("Revenue"),
• How much they spent on making those chips ("Cost of revenue"),
• How much money they spent on costs like R&D unrelated to the number of chips sold ("Operating expenses"),
• How much money they spent on or made from things like interest ("Other income"), and
• How much money they spent on taxes ("Income tax expense").

The reason the income statement does this is because it presupposes that the company is in a steady-state. It models the company as a machine that has inputs and outputs, and subtracting the latter from the former gives us profit. It doesn't account for "extraordinary" events like building a new factory, buying out a rival, or lending money to a supplier. Our problem is that in a growth industry, these events aren't extraordinary, so the income statement is misleading.

The key takeaway here is that there are ways to spend money that don't lower earnings. If earnings are all investors are looking at, you can sneak stuff past them.

A snapshot of investments

To find the investments, we need to look at a company's balance sheet and statement of cash flows. Let's start with the balance sheet:

The balance sheet is a snapshot of what Nvidia owned and what it owed in January 2025. The two sides that need to balance are the "Assets" and the "Liabilities and Shareholder's Equity". The sides have to balance because companies aren't free-standing, so to speak. Anything a company owns must also be owed either to creditors or to shareholders.

As an illustration, Nvidia had $8.6 billion dollars of cash. About $8.5 billion of this was owed to creditors as long-term debts. The remaining $0.1 billion is what the shareholders own as equity. The balance sheet just does this, but for every kind of asset and every kind of credit.

Keen-eyed readers will have noticed something weird here. The income statement said profit was about $73 billion, but the balance sheet shows a cash balance of only $8.6 billion (up from $7.3 billion the year before). There's an order of magnitude of difference. If the profit wasn't converted into cash, where did it all go?

There's two ways of answering this question: by looking at the balance sheet and by looking at the cash flow statement. Let's start by comparing the balance sheet entries across the two years.

We're looking for about $73 billion of difference. We see total assets went up from $66 to $111 billion, so there's $46 billion. Zooming in, we see the increase wasn't in cash which only went up by $1 billion. Instead it was in marketable securities ($16 billion), accounts receivable ($13 billion), inventory ($5 billion), deferred taxes ($5 billion), and property ($3 billion).

The increase in various [mostly non-cash] assets accounts for $46 billion of the $73 billion of profit. Where did the remaining $27 billion go? Nvidia disbursed it to shareholders and we can see this in the statement of shareholder's equity:

Zooming in on the third group of rows, we see that Nvidia repurchased some $34 billion of shares, paid $0.8 billion in dividends, paid its employees $4.8 billion of new shares, and paid some tax thing for $7 billion. This totals up to about $37 billion. This is about $10 billion more than were were looking for, but looking back at the balance sheet, we can see this came from an increase in liabilities.

So, of the $73 billion in profit, $37 billion were given to shareholders, $10 more billion were borrowed, leaving an increase in assets of $46 billion.

If Nvidia had done their $100 billion OpenAI investment last year, where would this have been visible? It would have decreased cash by $100 billion and created a new asset line called something like "Non-marketable securities" of $100 billion. Since Nvidia didn't actually have $100 billion of cash, they would've had to borrow it which would've increased one of the debt lines in the liabilities section.

The point to reiterate here is that Nvidia doesn't actually have $100 billion just lying around on their balance sheet to invest. In fact, Nvidia's total assets as of January were just $111 billion, so a $100 billion investment would literally double the size of the balance sheet. In theory, Nvidia could accumulate the necessary money by just not paying the shareholders $37 billion for three years, but that likely wouldn't work in practice because management generally like to hoard money, so they wouldn't be paying out so much unless they were being forced to.

That said, this is all fine in practice because the $100 billion OpenAI deal is actually just $10 billion of investment right now (archive link) plus a bunch of promises for the future. Increasing assets and liabilities by $10 billion seems very plausible on Nvidia's balance sheet.

Another thing worth noticing is that accounts receivable is $23 billion. In other words, a full third of Nvidia's profit comes from chips they've shipped to their customers, but which haven't actually been paid for yet. Looking at past years, this proportion of a third is par for the course, but it's still pretty surprising to see.

The takeaway here is that earnings don't necessarily mean more cash that can be freely spent. Also, the size of the balance sheet might be a lot smaller than earnings would have you believe.

Money spent on investments

Comparing the balance sheet entries for different years is a bit of a roundabout way to see what money was spent on. A more direct way is to look at the cash flow statement:

The cash flow statement lists all the changes to the company's cash account over the course of the year. The way to read it is to look at every line and ask what that line would do to a company's hypothetical bank account. This is a bit hard to conceptualize because many of the lines refer to year-on-year changes in the balance sheet entries, so their sign is counter-intuitive. Still, this is what we need to do to answer the question of how $73 billion of profit turned into only $1 billion of extra cash.

The cash flow statement starts with the net income which is theoretically how much more money the company made this year. This amount is then adjusted to come up with the actual cash Nvidia made this year from selling chips. For example, we know that accounts receivable increased by some $13 billion on the balance sheet from last year. This is money Nvidia's customers owe it, but haven't actually paid yet. Since this isn't cash, we subtract it from the theoretical profit. Note that the accounts receivable line on the balance sheet is actually $23 billion this year, but $10 billion of that was from last year and was accounted for in that year's cash flow statement. So, we only care about changes in the balance sheet lines, not absolute values. Once we've done these adjustments, we have cash flow from operating activities.

Next, we need to consider what investments Nvidia made. For example, we know that "Marketable securities" increased from $18.7 billion to $34.6 billion on the balance sheet. This means Nvidia spent about $16 billion of its profit to buy more securities. This is reflected in the cash flow statement by +$11 billion of securities that matured and became cash and -$27 billion of securities that were newly bought. Adding up all these, we get cash flow from investing activities.

Finally, we need to consider how Nvidia's financing decisions affected the cash balance. For example, we know from the shareholder's equity statement that Nvidia paid out $0.8 billion in dividends. This needs to be subtracted from the profit. We also know they issued another $0.5 billion in shares as employee stock compensation. This is money they spent and that was already subtracted from revenue in the income statement under operating expenses, but it's not actually cash. Since this expense isn't in cash form, it doesn't decrease the company's cash balance, so we need to add it back in the cash flow statement. Adding up all of these, we get cash flow from financing activities.

Note that the section totals are independent, so we're not actually working down the page. Rather, we're computing three totals which will be added together at the end. For the year ending January 2025, Nvidia made $64 billion in cash from its operating activity of selling chips. It spent $20 billion on investing activities. And it spent a further $42 billion on financing, mostly on buying back shares. This leaves about $1 billion which is the increase to the cash position.

Where would the $100 billion OpenAI investment go? It would clearly appear as a line in the investing activities section and it would increase the total of that from -$20 billion to -$120 billion, a 500% increase. This is assuming they'd actually pay out the full amount, which we know they didn't do. They only paid $10 billion, so it would only increase investing activities by 50%.

An interesting thing to note is that building new chip foundries would also have fallen under investing activities. I assume this is the "Acquisitions, net of cash acquired" line of $1 billion. That does not seem like a large number to me in the context of chip foundries (to give a sense of scale, the CHIPS act had some $280 billion of subsidies for semiconductor manufacturing). It seems like Nvidia is doing a lot more investing into marketable securities than into increasing capacity.

The strength and weakness of the cash flow statement is that it talks strictly about what happened to the company's cash position over the year. I've mentioned the $23 billion of accounts receivable which is money Nvidia is owed, but hasn't yet received from its customers. They are probably going to get most of that money, so it's a bit odd to exclude it.

The takeaway here is that the cash flow statement is where we find out how money was spent. However, it is not enough to paint a full picture. The income statement, balance sheet, statement of shareholder equity, and cash flow statement fit together like puzzle pieces. Looking at any in isolation is misleading. All need to be read at once.

Boosting revenue

We've seen how the financial statements fit together. Now let's do some financial engineering.

Let's imagine we're Jensen Huang, CEO of Nvidia, and it's January 2025. We've just ended the financial year with the above statements. It's been a pretty good year with profits growing to $73 billion.

In fact, this is so good we don't know what to do with the money. We're not spending it on building new chip foundries, presumably because we have or can source all the capacity we need. We increased R&D by $4 billion, but there's a limit to how much money we can throw at that. We had so much money that we just returned half of it to shareholders in the form of share buy-backs. The rest, we dumped into buying marketable securities.

Our fiduciary duty as CEO of Nvidia is to increase shareholder value. For a company that sells chips like Nvidia, the main way to do this is to sell more chips. The obvious problem is that our customers, the AI companies, aren't actually making a lot of money. Or any money, really. Basically all of them are money furnaces that are just burning investor money on building out datacenters without actually bringing in much revenue.

As CEO of Nvidia, we know how to make chips, but we don't know much about AI. All of our customers, the AI companies, seem to be sure that if only they could 10x their capital investments, then they'd be profitable. We want our customers to be successful, so that they can afford to buy more chips. However, just OpenAI alone wants to spend $1 trillion on new datacenters. That's 13 years worth of Nvidia profits, so Nvidia can't do this alone. But maybe it doesn't have to.

The traditional vendor financing scheme would look like this:

• Nvidia has $100 billion of cash saved up and wants to boost next year's earnings.
• Nvidia uses this $100 billion of cash to invest in or loan to OpenAI. This is an operation that happens off the income statement, so it has no impact on Nvidia's revenue or earnings. Instead it only appears on the balance sheet and cash flow statement.
• OpenAI receives this investment or loan. Since it's not an operational event, it also doesn't affect OpenAI's income statement. It just appears on the balance sheet and cash flow statement.
• OpenAI turns around and buys $80 billion of Nvidia chips. Nvidia records this as revenue next year and given their margin, sees earnings increase by some $40 billion.
• Investors are happy that Nvidia is showing earnings growth, so they buy the stock and drive up its price. Management are happy because they get paid in stock which is now worth more.

The problem with the traditional vendor financing scheme is that it's lossy. Nvidia spends $100 billion of cash to get back $80 billion of revenue. You can do this for a few years, but eventually you run out of cash reserves and the party stops.

The way to keep this going is to get other people to buy in. The dream scenario would be for Nvidia to spend only $10 billion of cash on OpenAI and convince others to put up the remaining $90 billion, so that Nvidia's revenue next year still increases by $80 billion. However, I think this is pretty hard to pull of with just two companies because it's pretty obvious what's going on and investors aren't stupid enough to fall for it (after they got burned by this scheme in several previous bubbles).

But what if you have more companies? Consider this scheme:

• Nvidia invests or loans $100 billion of cash into OpenAI. This happens off income statement.
• OpenAI spends $80 billion on ordering new datacenters from Oracle. This is again off income statement for OpenAI.
• Oracle records $80 billion in revenue and spends $60 billion on ordering chips from Nvidia.
• Nvidia sees $60 billion of new revenue.

At first glance, this is worse for Nvidia because they spent $100 billion to make back only $60 billion. However, the total revenue in the AI ecosystem has increased by 80+60 = 140 billion. We can make the chain longer and increase total ecosystem revenue further. If Oracle didn't build the datacenters themselves and just farmed out the work to a subcontractor for $60 billion and that subcontractor bought $40 billion of Nvidia chips, then the total revenue would be 80+60+40 = 180 billion.

Vendor financing chains like this don't make Nvidia more money, but the more revenue the AI ecosystem shows, the easier it is for the companies to get more investor money. This way, Nvidia wouldn't have to spend the full $100 billion and could actually turn a profit.

To recap, the way Nvidia makes more money is for the AI party to keep going. The problem is that it isn't going very well because the AI companies aren't profitable. Nvidia can't make the AI companies better, but it can trust them when they say they need more investor money, and Nvidia can help make them look better to investors.

In the Owenomics article linked above, the author argues that we're not in an AI bubble because there isn't enough outside investment flowing into the ecosystem. Instead, we're in an AI boom because it's established players redirecting their investment budgets into this. I agree, but I think the problem is that OpenAI needs a bubble in order to raise the money to spend on their $1 trillion build-out. And all the other AI companies are in a similar situation—they're not bringing in enough revenue, so they're going to need an order of magnitude more money than the established players can afford to spend. For this to happen, they're going to need to look good to new investors. Nvidia understands this and they're trying to make it happen with chains of vendor financing.

What's so off-putting about this

When I first heard about this circular financing setup, I think I had the same reaction as most people and was put off by it. The specific thing I didn't like was that this makes all the companies harder to reason about.

In a world ideally suited for company valuation, Nvidia would make GPUs and OpenAI would make AI—each company would be a pure play in their area of expertise. But as a result of all these schemes, they're now entangled in complicated ways. For example, we know Nvidia are good at making chips and we can model how wide their moat is. But now we also have to wonder how savvy of an investor Jensen Huang was when he put money into OpenAI.

The more connections there are, the harder it is for any one company to be considered in isolation. The more connections there are, the harder it is to invest in just one company and not the entire sector. After all, if I wanted to invest in the sector, I'd buy an ETF or a specialist fund.