OpenAI Must Scale a Massive Money Mountain
Cory Weinberg
The topline numbers were already pretty staggering, now Cory Weinberg of The Information has far more granular detail on how OpenAI is projecting their new several years to play out from a business perspective. And they're... pretty wild!
We already knew the big goal of reaching $100B in revenue by 2029 (and my walk up to that number – with the help of ChatGPT, naturally – ended up being pretty close). But ChatGPT itself would go from a projected $3B business this year to a nearly $40B business in 2028. That would make it roughly the same size as the Mac business is today for Apple. Bigger than all of Qualcomm, Starbucks, and Uber. In 2029, they're projecting $55B. That would put ChatGPT ahead of where Oracle, Nike, and Intel are today. Just ChatGPT.
The API business model is fairly interesting in that it grows pretty slowly until eventually pushing back to nearly the quarter of the overall business that it is right now by 2029. The "Other Products" line, which does not exist today but grows to a nearly $25B business in 2029. I mean... literally who knows?
It isn’t clear what those new products might be, but the company is currently pursuing products involving agents that can use a person’s computer to handle complex and monotonous tasks, as well as a research assistant, people with knowledge of the company said. And it has discussed selling higher-priced subscriptions to its most advanced technology. Other products that haven’t fully hit the market include its Sora video generator, a more-direct competitor to Google Search, and software for developers of robots.
It's certainly one way to model up to that nice, even $100B though.
On the flip side, the projected losses to get there will be massive:
Before it gets to that point, losses could rise as high as $14 billion in 2026, nearly triple this year’s expected loss, according to an analysis of data contained in OpenAI financial documents viewed by The Information. This estimate doesn’t include stock compensation, which is one of OpenAI’s biggest expenses, although not one it pays in cash.
That $14B loss is largely because model training cost are projected to get close to $10B that year. And the compute costs grow from there. The good news, I guess, is that those costs should be tied to revenue growth (though what does, say, the Apple deal do to this ratio?):
OpenAI expects to spend more than $200 billion through the end of the decade, excluding stock compensation costs. Between 60% to 80% of its spending each year would go toward either training or running the models.
One of the biggest things any investor in this most recent round would have had to get comfortable with is not just that OpenAI will need to raise more money after this massive fundraise, they're going to have to raise a lot more money.
An analysis of the documents suggests that OpenAI projects its total losses between 2023 and 2028, excluding stock compensation, will be $44 billion. The same analysis suggests the company anticipates it will make $14 billion in profit on that basis in 2029.
Per the reporting, OpenAI had about $1B in cash on the balance sheet prior to this round. So that gives them roughly $7.5B now – but we don't know how much of the $6.6B round will ultimately be in the form of compute credits from Microsoft. Still it's roughly the same as it's cash they would otherwise have to send to their partner to operate.
Those cloud credits seemingly make it hard to calculate actual burn though. Previous reports pegged it at $5B last year. This report has it quite a bit lower the first half of the year – but again, likely with credits taken into account. If we're going to assume some of the fundraise is in credits and not cash, rough math would indicate that they're going to need something just north of $30B to operate between 2025 - 2028. So let's call it a nice even $25B more than they have right now in terms of cash to burn.
OpenAI is clearly very good at fundraising. Understatement of the century, thus far? But a lot can happen between now and 2028. An in AI, where the days feel more like months, even more is unknown. That's a massive risk, to put it lightly.
Of course, they'll have levers they can pull to dial down burn, in particular with compute. But there's also a chance they'd have to dial it up as well. It's just a complete unknown at this point. It does feels like it will be hard for any other startup to compete with such cash needs – Anthropic, answer your phone, Amazon is calling – but OpenAI will be competing with Google, Meta, and all the rest of Big Tech. Maybe not Apple. But probably them too, eventually.
But Microsoft is in their corner, right? Right?! I mean, sort of! They're in OpenAI's corner but there are at least three other corners in their square logo. Actually, there are far more corners in the way they do the logo. I'm stretching it. But maybe not as much as Microsoft in trying to hedge this particular bet. Point is, I'm just not sure how much we can count on Microsoft to backstop OpenAI all the way to 2029. Then again, given their stake in eventual profits, maybe they will want to keep the dice rolling until 2029 – if that ends up being the actual timeline for profitability.1 In my experience, things tend to not play out to such projections. Crazy, I know. Then again – again – the shift to for-profit status may or may not upend the whole profit-sharing arrangement, so who knows?
Maybe OpenAI can entice Apple back to the table before then. It probably depends on how Apple's own AI efforts go. If things are looking a bit dicey post-launch, perhaps they turn back to their partner for a deeper partnership. For real, this time. Otherwise, SoftBank is undoubtedly willing to double, triple, and quadruple down – unless they're also in dire straits, as tends to happen to the entity every so often. So it might mean turning to more sovereign wealth funds.
Or, god forbid, going public. But the question there is similar with regard to timing. It's a question of not only the IPO window being open, but the public markets willing to tolerate a massively money-losing company. Such appetites also ebb and flow. Right now, profits are paramount. Well, unless you're Donald Trump's social media company – or maybe an AI company! Though even those tend to be profitable in the public markets right now. So...
1 As noted in the report, OpenAI is projecting showing a profit in 2026, but that seems to be using accounting that amortizes the cost of training, spreading it over years. But they'll still need the cash upfront to spend on it...