They Don't Have the Money: OpenAI Edition

Tl;dr another installment of They Don’t Have the Money (previous edition: Oracle)

OpenAI’s Wile E. Coyote moment approaches.

OpenAI is the generational company of generative AI (sorry Meta, sorry Elon, it isn’t even close). They defined the space and lead the market, combining enormous ambition with stratospheric spending plans, unironically projecting needs in the trillions of dollars.

OpenAI is also patient zero in any epidemiological diagnosis of the AI bubble. What happens if the market leader can’t make the economics work? What happens to the ever-growing parade of vendors lined up to take OpenAI’s money (and pump their stocks)?

They’re everyone’s most sought after customer (except for Microsoft who walked away from providing OpenAI’s training infrastructure, a move perhaps warranting broader contemplation).

Hardly a day goes by without OpenAI announcing plans to spend billions and billions. $300 billion for Oracle! $22.4 billion for CoreWeave! $10 billion for Broadcom!

And there are deals denominated in other (obscuring) currencies. They’re buying ten gigawatts from NIVIDA! Six gigawatts from AMD! 900,000 wafer starts a month from Korea! 230 Norwegian megawatts! 200 UAE megawatts!

The FT estimates they’ve “signed about $1tn in deals this year for computing power”.

They haven’t even started to discuss the costs of new power generation to provide juice for all that compute:

“The deals would give OpenAI access to more than 20 gigawatts of computing capacity, roughly equivalent to the power from 20 nuclear reactors, over the next decade.”

Energy is the long pole, with both costs and delivery times soaring. A new gas turbine now takes as long as seven years. The last nuclear plant built in the US took 11 years and cost over $30 billion (I’ll be delighted if they can improve on that). So OpenAI isn’t done announcing enormous checks.

Beyond compute the company has other non-trivial cash expenses. Meta’s AI mulligan has driven AI researcher salaries up even further. Jony Ive has “15 to 20 ideas” for notoriously capital-intensive hardware. There are startup investments. And who knows what else.

The burn rate is scorching, with CEO Sam Altman saying OpenAI might be the “most capital intensive” company of all time.

The projected cash burn keeps growing. The latest is $115 billion through 2029: “That’s $80 billion more than it forecast in the first quarter”. That was like a month ago, before the latest slew of commitments.

Maybe the Real AGI is a Perpetual Money Machine

It may seem like nitpicking, but massive spending commitments eventually require massive amounts of actual cash. Either OpenAI raises a boatload of money from investors, and/or they build cash flow positive businesses that can organically fund hundreds of billions of investments.

They’re coming up light on the fundraising side, and it is a little puzzling that investors aren’t throwing money at the premier AI company. Instead we’re seeing balance sheet games, vendor financing, and suspiciously round and bombastic headline numbers.

Do investors dislike the terms on offer? Are they put off by OpenAI’s corporate structure (or lack thereof)? Do the financial projections not pencil out? (This many zeros takes us well past vibes-based investing). Are there hallucinations in the data room?

OpenAI is already the most valuable startup the world has seen with a $500 billion valuation, and might be on track to have cumulatively raised $60 billion by the end of this year. Valuation alone makes it hard to raise more. But even if they were to raise $100 or $200 billion on a trillion-dollar valuation, it still won’t meet their cash needs.

The Stargate announcement of a few months ago claimed $100 billion would be “immediately” put to use with a headline number of $500 billion. Someone said to Bloomberg at the time “the headline $500 billion number is science fiction, just like the movie it is named after”. Stargate soon dissipated into a fuzzy ingredient brand, and even Softbank’s $30 billion has been delayed while OpenAI gets its corporate structure figured out.

Someone also suggested Stargate was “turning into a party round of both cash and electricity. Raising in increments of billions and megawatts is the long road to trillions and gigawatts.”

The recent NVIDIA investment in OpenAI is a $100 billion headline number but just $10 billion up front (and unlocking the next $10 billion requires OpenAI to spend $50+ billion, per NVIDIA). In other words, the Stargate “immediately” sum is now the aspirational headline number. Near term funding is actually receding, while commitments boom.

Central banks act as lenders of last resort during financial panics. That NVIDIA has (already) become the funder of last resort for OpenAI (and most of the neoclouds) isn’t an auspicious sign.

Sam Altman describes the $100 billion headline number as only putting “a small dent in it”. OpenAI President Greg Brockman says “We’re still three orders of magnitude off of where we need to be.”

Statements like “I don’t think we figured out yet the final form of what financing for compute looks like,” and “I suspect we can design a very interesting new kind of financial instrument for finance and compute that the world has not yet figured it out,” suggest neither OpenAI nor GPT-5 have yet come up with a way to outwit the gravitational laws of finance.

A Secret Plan?

Sam says they do have a (as yet secret) plan, but gives no clues of where a huge cash injection might come from:

Give us a few months and it’ll all make sense and we’ll be able to talk about the whole — we are not as crazy as it seems. There is a plan.

…when we are done with this new set of deals and get to explain here’s where we’re going and why, I think it will all make sense. People may say it’s too ambitious, it’s too risky, whatever. But at least why we’re doing the stack that we’re doing of stuff I think will make sense.

[Q.] This is all amazing, are we going to have enough electricity for all this?

…please give me the grace of a few more months to finish. I promise to tie the whole thing together.

Hopefully the AGI financing plan is more than just warrants for 10% of AMD.

I tried to help OpenAI raise some money by starting a GoFundMe for them, but those fundraising campaigns are capped at a mere billion dollars. Which doesn’t even pay for snacks at OpenAI.

“No Wall Left Behind”

Sam suggests they will pay for their spending needs from revenue. OpenAI has built an incredible business. They’re the largest AI company by far (and already one of the biggest software companies), over $12B in ARR and 800 million MAUs, with unprecedented growth at their size.

They’re the leading AI model lab, but also have become a very good app company, with great product velocity. Their strategy is to throw a ton of stuff at the wall to see what sticks.

But many of those apps being flung at the wall are thin, arguably optimized for headlines, as opposed to actually wresting franchises away from the biggest companies in technology, which seems to be their target. Tremendous TAM, but also a daunting degree of difficulty. Amazon, Apple, Google, Meta, Microsoft, LinkedIn, Tik Tok, amongst others, aren’t going to easily surrender their franchises to vibe-coded challengers. And frontal assaults on tech’s biggest franchises bode poorly for OpenAI’s sorely needed pricing power or margins.

They need massive cash flow to pay their own way. These apps aren’t traditional software businesses with negligible marginal costs. AI has immense training costs and hefty inference COGS. A risk is OpenAI spreads themselves too thin, and doesn’t create any giant profit pools of its own.

Supermegaloblitzscaling

I’m not even remotely an AI doomer, Luddite, or degrowther (nor European, but then I repeat myself), and defer to no one in my enthusiasm for CAPEX spending.

But as it currently stands, OpenAI needs hundreds of billions of dollars injected into the system to have a chance to realize their dream. Maybe they can stick the landing (and I’d welcome it), but it isn’t the way to bet today.

In the absence of that massive injection of cash, something has to give. Infrastructure budgets. Price points. The freemium model. The breadth of ambition. Aluminium. Component prices. Non-GAAP RPO fantasies. Algorithmic inefficiency.

The Wile E. Coyote moment occurs when the coyote runs off a cliff, but remains suspended in midair (no doubt preoccupied with feverish visions of AGI), not noticing just how far below lies the floor of actual funding. Eventually, our coyote glances down, common knowledge crystalizes that they don’t have the money, and gravity takes over.

Let me know in the comments if I’ve somehow overlooked a few hundred billion in cash somewhere.