OpenAI burned $34B to earn $13B — and the thread can't explain $5.7B of marketing

What the leaked audit actually shows

A set of audited financials for 2025, reported by Ars Technica and corroborated across outlets, puts hard numbers on a question the industry has been circling for two years: how much does it cost to run OpenAI? Revenue more than tripled to $13.07 billion. Total costs and expenses hit $34 billion. The loss from operations widened from $8.78B in 2024 to $20.92 billion in 2025.

The headline "billions in losses" hides an important caveat. A $41.55 billion charge tied to the for-profit conversion — a change in the fair value of convertible interests and warrant liabilities — is a non-cash accounting entry, not money leaving the building. Strip out the non-cash and one-off charges and the real out-of-pocket burn for the year lands around $8 billion. Large, but not the apocalyptic figure the raw loss line implies.

The $5.73B nobody in the thread could account for

R&D at $19B and compute at $7.5B surprised no one. The sales-and-marketing line did. "I'm a simple guy and I don't understand the 'sales and marketing' cost," wrote an engineer puzzling over the number in the thread. "To the extent they work and there is a customer base what marketing could you possibly be engaged in? Doesn't the product sort of market itself?"

The replies split into three theories. The mundane one came from a commenter who'd seen the spend in person:

"I've seen physical billboards in the Portland, OR area for OpenAI… Not really sure what kind of return they're getting on those but apparently they can just do whatever they want, even if they're losing money."

A more structural read came from a commenter pointing at the enterprise motion:

"They have a large and rapidly growing enterprise sales organization. If you want to sell to enterprises you need account executives, solutions engineers, forward deployed engineers, etc."

And Datasette creator Simon Willison flagged that the numbers had already surfaced a day earlier — this is a story the community has been chewing on, not a one-day shock.

Why a category-defining product still needs to buy attention

The sharpest explanation reframed the whole question: the moat is thin. One commenter put it plainly:

"They need marketing because they have competition that essentially offers an identical product. Why should a consumer choose openai over anthropic or whatever else there is? The answer is not obvious."

A darker theory pointed to influence operations — a commenter linked to reporting on a super PAC backed by OpenAI that pays creators to praise US models and discredit Chinese ones. Others noted that the consumer subscription is heavily discounted against API pricing, and a deep-learning engineer suggested that gap, plus a string of $50M grant initiatives, could be booked partly as marketing.

The real question the thread kept asking: how much will you actually pay?

Underneath the cost analysis ran a referendum on monetization. One commenter asked how far people would go — double the price? triple? — before walking. The answers map the gap OpenAI has to close. One user said LLMs were worth as much to him as a car and he'd pay "possibly a thousand a month." Most said the opposite:

"Max 60 bucks a month. More than that and I'd just move to local qwen 35b or some other cheaper model on openrouter." — a commenter weighing the local-model alternative

That is the squint worth holding onto. The losses are survivable as long as the capital keeps flowing, and the non-cash charges make the headline number scarier than the cash reality. But the willingness-to-pay ceiling in the thread sits at $30–$60 a month for individuals, with a hard exit toward cheaper Chinese and local models the moment prices rise — exactly the substitutability that forces a $5.73B marketing bill in the first place.