The Art of the OpenAI Deal
By
While they published this on Friday – before the subsequent WSJ report that Apple had dropped out of the financing – it's worth digging a bit more into the numbers that Mike Isaac and Erin Griffith obtained with regard to OpenAI:
OpenAI’s monthly revenue hit $300 million in August, up 1,700 percent since the beginning of 2023, and the company expects about $3.7 billion in annual sales this year, according to financial documents reviewed by The New York Times. OpenAI estimates that its revenue will balloon to $11.6 billion next year.
But it expects to lose roughly $5 billion this year after paying for costs related to running its services and other expenses like employee salaries and office rent, according to an analysis by a financial professional who has also reviewed the documents. Those numbers do not include paying out equity-based compensation to employees, among several large expenses not fully explained in the documents.
While it may not be normal for a public company not to include stock-based compensation in their earnings,1 it is normal for a startup to exclude this in any documents shared with investors. We can debate if a company being valued at $150B is still a "startup", but it's not a public company, so, yeah, right or wrong, that's the standard in such information packages. Though given who OpenAI has to talk to in order to raise these sums, certain investors here may not like that too much. Like, say, Apple?
This will likely come even more into play as OpenAI shifts to a for-profit entity and equity conversion becomes a quagmire that will make the current chaotic dynamics of the company seem like child's play. If the company had to/were to report stock-based compensation as a part of their bottom-line – that's going to make the $5B loss seem pretty small. But again, this has long been a debate (see: footnotes below).
Without question, it's incredible revenue growth at a massive scale. Per the NYT numbers, here's the revenue climb:
- 2024:2 $3.7B
- 2025: $11.6B
- ...
- 2029: $100B
Naturally, I asked ChatGPT to fill in the blanks. FWIW, o1 – "Strawberry" – seemed to get stuck "thinking" about it. But old, "trusty" 4o gave me the following:
- 2026: $18B
- 2027: $32B
- 2028: $56.5B
In other words, this is assuming a roughly 77% growth rate annually as it approaches $100B. But that's undoubtedly just ChatGPT backing into filling in those blanks; one has to assume the growth rate is going to slow approaching those numbers. So I asked ChatGPT to be a bit more realistic with a decaying growth rate – here's what I got:
- 2026: $28.5B
- 2027: $55B
- 2028: $84B
That leaves us at just over a 19% y/y growth rate from 2028 to the $100B 2029 year. Seems more reasonable. But who knows, it's all just guesstimates – especially in the world of AI. Revenue could grow even faster with new opportunities in the next five years, or it could fall – the horror – in the face of competition. It doesn't really matter right now, this is just me trying to wrap my head around the numbers reported and doing some math – well, having ChatGPT do some math, thus adding to OpenAI's expenses. Moving on...
If the company is raising $6B - $7B and burning $5B, that obviously doesn't buy them a ton of extra time on top of whatever money is in the bank right now. But this would seem to suggest that they believe those expenses are not going to keep scaling in line with said revenue growth. Again, you'd hope that's the case – obviously – but, well, there are a lot of unknowns here.
The documents offer the first detailed look into OpenAI’s financial performance and how it is presenting itself to investors, but they do not neatly explain how much money it is losing. The fund-raising material also signaled that OpenAI would need to continue raising money over the next year because its expenses grew in tandem with the number of people using its products.
OpenAI declined to comment on the documents.
OpenAI’s revenue in August more than tripled from a year ago, according to the documents, and about 350 million people — up from around 100 million in March — used its services each month as of June.
Most of that has come from the continuing popularity of ChatGPT, which was released in November 2022. The documents show a spike in growth after ChatGPT began allowing people to use the service without creating an account or logging in. The company expects ChatGPT to bring in $2.7 billion in revenue this year, up from $700 million in 2023, with $1 billion coming from other businesses using its technology.
Here's an obvious example of a big unknown: ChatGPT is on the verge of rolling out to millions of Apple Intelligences users on iOS, iPadOS, and macOS later this year. It would seem that those users, for the most part, are also going to be using the service without being logged in. This remains one of the biggest question marks for me around all of this – while the initial reporting had Apple not paying OpenAI for such usage, you can now get a sense from these numbers just how expensive such usage is likely to be for the company. It's extra awkward because it's money they're mostly going to have to pay Microsoft, for the servers on which OpenAI is running. So OpenAI is going to be paying Microsoft for Apple's usage of ChatGPT. And while it made sense that Apple would be investing here, thus giving OpenAI money to pay Microsoft in that transaction. Now that they're seemingly not, it's going to be OpenAI giving other investors' money to Microsoft for Apple's usage. And one of those other investors is Microsoft! So Microsoft could be giving OpenAI money to pay back to themselves for Apple's usage.
So that's fun.
Now, perhaps Microsoft is actually just "investing" cloud credits here, as has largely been the case with the $13B "investment" to date. But still. It's a weird situation. And an expensive one. To the point where it feels like there simply must be a missing piece of these equations here...
OpenAI is offering unusual deal structures to investors. Thrive Capital has invested $750 million into OpenAI’s latest round of funding, according to a person familiar with the deal. In addition to putting in its own money, the firm plans to use a financial instrument called a special purpose vehicle to gather an additional $450 million from other investors, the person said.
As the deal’s lead investor, Thrive also has an unusual perk: the option to invest up to $1 billion more into OpenAI at the same $150 billion valuation through 2025, according to the documents. That could be lucrative to Thrive given how quickly OpenAI’s valuation has escalated to $150 billion, up from just $30 billion a year ago.
None of OpenAI’s other investors have been granted the same terms, and some of them were frustrated by the special preference, according to two people familiar with those discussions.
Yeah, that's not that unusual. It's a warrant. It became a bit more unusual during the go-go era of investing the past decade, but at the later stages, these kinds of "sweeteners" are still used – in particular for deal leads, like Thrive here. It's a bet that the price (valuation) will keep rising and so it would effectively make this a much better deal for the firm if that happens – and if history is the guide here, this is what has happened just about every year to date. Warrants are also often used if the company really wants to get someone else to invest. Like, say, Apple?
(Okay, I'll stop now. It just strikes me a particular problematic for them to back out of this deal!)
One more thing: a decidedly less common deal term...
As part of the investment round, OpenAI has two years to convert to a for-profit business, or its funding will convert into debt, according to the deal documents.
Convertible notes, of course, are very common. But this is essentially the opposite. Whereas a typical note will convert debt into equity at a latter date depending on certain terms met, this is money meant to buy a piece of the company – that is, equity – turning into debt that is to be repaid. As weird as that seems, it perhaps makes more sense here given that money from investors isn't actually used to buy equity in OpenAI, since the current structure doesn't allow for that – you only are buying the rights to (capped) profits.3 Whereas the warrants are upside, this is a very real downside protection.
And it seems like it could be a very real problem for the company come 2026 if it's neither a for-profit nor insanely profitable... Clearly they're hoping for the former but believe the latter will protect them if something goes awry.
Something going awry, at OpenAI? Come on...
M.G. Siegler
M.G. Siegler
M.G. Siegler
M.G. Siegler
M.G. Siegler
1 Though many try to obfuscate these numbers via non-GAAP reporting when possible as it has basically always been a debate as just how much of an "expense" this actually is since no actual money changes hands (instead it's diluting the equity value of current shareholders).
2 ChatGPT itself is said to account for $2.7B of the $3.7B in 2024 revenue, with the remaining $1B coming from "other businesses using its technology".
3 I'd be curious to know if you keep the rights to the profits or if the conversion to debt also converts those stakes...