Leaked financial documents reveal that OpenAI posted a $38.5 billion net loss in 2025, despite pulling in $13 billion in revenue. The company spent $34 billion in total, roughly $17 billion of which went directly to Microsoft Azure, because when you’re building the future of intelligence, the electric bill is apparently the present of bankruptcy.

For context, the loss jumped from $5 billion in 2024 to nearly eight times that in a single year. The operational loss alone ballooned from $8.78 billion to $20.92 billion. These audited figures surfaced ahead of OpenAI’s planned IPO, which the company hopes to complete by December 31, 2026 — presumably before anyone runs the numbers twice.

🤚 The Open-Palm Audit

Let’s appreciate the sheer ambition of losing $38.5 billion in a calendar year. This is not a rounding error. This is not a write-down. This is a company that made $13 billion — a figure most enterprises would celebrate with champagne and a press release — and then spent $34 billion like a teenager with their first credit card and a deadline.

The breakdown is instructive:

• $17 billion to Microsoft Azure for compute — making OpenAI simultaneously Microsoft’s largest customer and most expensive investment
• $20.92 billion in operational losses — up from $8.78 billion the prior year, because scaling has a learning curve and the curve costs $12 billion
• $13 billion in revenue — a number that would make any normal company profitable and makes OpenAI technically a charity with excellent marketing

Sam Altman’s company is now the most expensive loss-making enterprise in human history that isn’t a government.

👐 The Two-Handed Balance Sheet

The real story here isn’t the loss — it’s the trajectory. OpenAI’s losses didn’t merely grow; they performed a gymnastics routine that would earn a 9.8 from the Romanian judge. From $5 billion to $38.5 billion in twelve months is the kind of exponential curve that Sam Altman usually reserves for talking about intelligence itself, not his company’s deficit.

And yet, the IPO marches forward. This is the financial equivalent of a man on fire walking into a bank and asking for a mortgage. “Yes, I am currently ablaze. But have you seen my growth metrics?”

The Azure dependency is particularly poetic. OpenAI pays Microsoft $17 billion a year. Microsoft invested $13 billion in OpenAI. OpenAI’s entire revenue is $13 billion. Microsoft is, in effect, funding a company whose primary expense is paying Microsoft. This is not a business model. This is a Möbius strip with a billing address.

The planned IPO by end of 2026 adds urgency to a situation that already had the emotional texture of defusing a bomb in a hurricane. Public markets will see these numbers. Retail investors will do math. And math, unlike GPT-5.5, does not hallucinate — though it may cause hallucinations in anyone holding the stock.

🌿 The Gentle Awakening

There is something almost philosophical about a company that builds intelligence losing money at a rate that suggests the opposite. OpenAI has created models that can write poetry, debug code, and pass the bar exam, yet its own balance sheet reads like a cry for help written in red ink.

Every dollar of revenue is accompanied by nearly three dollars of spending, which means OpenAI’s unit economics are currently “sell a dollar, spend three, and hope that superintelligence arrives before the accountants do.”

And perhaps that’s the bet. Perhaps the argument is that once you build a god, the god pays for itself. But in the meantime, the god requires $17 billion in Azure credits, a planned IPO, and the collective faith of investors who believe that losing $38.5 billion is simply the cost of doing unprecedented business.

👑 The Gold-Leaf Reckoning

The IPO filing transforms these numbers from an internal embarrassment into a public invitation. Soon, ordinary investors will be asked to buy shares in a company that loses money faster than it makes it, whose largest expense is paying its largest investor, and whose path to profitability requires inventing a new kind of intelligence and then selling it to everyone on Earth.

This is either the most ambitious business plan in history or the most expensive science experiment ever conducted outside of CERN. The difference between the two scenarios is about $300 billion in market cap and a lot of nervous laughter in board meetings.

One thing is certain: when OpenAI goes public, the S-1 filing will be the most widely read horror novel of the fiscal year.

“We lost $38.5 billion, but we’d like you to know it was audited, which means a professional accountant looked at the number, confirmed it was real, and then presumably sat in their car for a while.” — The Slap of Wisdom Financial Crimes Desk, calculating the per-token cost of existential dread