Audited: OpenAI burned $34 billion. The plan? Burn another $122 billion.

No, that's not the GDP of a small country. That's the burn rate of a company selling $20/month subscriptions.

Here's the breakdown, for the number lovers:

• $19 billion in R&D — aka "we're still trying to make GPT-5 work without it collapsing"
• $6 billion in Sales & Marketing — because obviously when you have 900 million weekly users, you need another $6 billion to convince them to use the product they already use
• And the other $9 billion? Probably catering for the meetings where they discuss how to cut costs.

The best part? These are audited figures. Not analyst estimates, not optimistic projections. Real numbers, signed off by someone holding a trembling pen.

And the result of all this? GPT-4o that sometimes remembers, sometimes doesn't. O1 that reasons well but costs more than a McKinsey consulting engagement. An API that burns ~67 cents in inference for every dollar of revenue.

But wait, it gets better.

OpenAI projects $14 billion in losses for 2026, with training costs rising to $32 billion. Profitability? Forecasted for 2029-2030. Maybe. If everything goes well. If another DeepSeek doesn't emerge to humiliate them with a model trained for the price of a Milan apartment.

And gross margin? 33%. For a software company. Thirty-three percent.

Context: Salesforce runs 75%. Adobe, ~80%. Even Uber — which for years burned cash like it was confetti — is now profitable. But OpenAI? No, OpenAI is "different." It's "infrastructure." It's "the future".

The future costs $34 billion a year and isn't sure if it exists tomorrow.

And the $6 billion in marketing? I'm picturing YouTube ads with Scarlett Johansson saying "Hi, I'm the AI" (oops, already did that), Apple partnerships that generate no direct revenue, and banners on every tech site on the planet. Because when you have a product that sells itself, clearly you need to spend $6 billion explaining why people should pay you $20 a month.

The market: "Wow, $25B ARR! Extraordinary growth!"

The reality: "Yeah, but for every dollar coming in, two go out. And we're raising another $122 billion because the first $34B wasn't enough".

And the CEO? Busy explaining that "inference cost scalability will improve." How? Unknown. When? Unknown. But trust me, bro.

Meanwhile, Anthropic burns "only" $19 billion, has better margins, and Claude Code is stealing market share in coding.

Google DeepMind? Profitable yesterday, thanks to Cloud revenue.

But OpenAI has ChatGPT. And 900 million users. Of which only 5.5% pay.

So yes, $34 billion spent to monetize less than 6% of the user base. A commercial efficiency that would make even WeWork blush.

And the plan? "Super app." Translation: "Our core product isn't profitable enough, so let's build an app that does everything".

Because when you can't figure out how to make a chatbot profitable, the obvious solution is to add taxis, food delivery, and payments. Worked so well for... oh right. Nobody.

$34 billion. Audited. And next time someone tells you AI is "democratizing" anything, remember: it's mainly democratizing how venture capitalists burn pension funds.