Let me give you three numbers and one question.

$19B ARR. That's OpenAI's current annualized revenue. Real money, growing fast. ChatGPT Plus subscribers, API customers, enterprise contracts — it adds up.

$14B. That's what they burn every year. Compute, talent, research, infrastructure. They are spending $14 billion to generate $19 billion. The gross margin story is not pretty.

$840B. That's the IPO valuation target. At $19B ARR, that's a 44x revenue multiple. For context, Salesforce trades at roughly 6x. Nvidia, at the height of the AI boom, was around 30x. OpenAI is pricing itself like it will be the most important company in history.

Here's the question: what bridges those three numbers?

What They're Doing About It

Two moves this week tell you the strategy.

Sora is dead. The $1B Disney deal is terminated. The video generation product is being wound down. This is not a pivot — it's an amputation. Sora was burning compute without a clear monetization path. At $14B annual burn, you cannot afford products that are spectacularly impressive but commercially vague. The mouse is no longer involved.

Ads are live. OpenAI launched a pilot advertising program and hit $100M ARR in six weeks. Six weeks. That's not a product — that's a signal. The signal is: we have 600 million monthly active users and we are finally going to monetize their attention directly. If the ad product scales, it changes the burn math faster than any enterprise deal could.

The Honest Tension

Here's what the ad move costs them: the trust of the developer community that built on the promise of "we're a research lab, not an ad platform." Anthropic will absolutely use this in enterprise sales. "We don't sell your attention. We sell compute." That's a clean line.

And here's what killing Sora costs them: the narrative. OpenAI was the company that could do everything — text, code, image, video. Pulling video shrinks the TAM story right before an IPO. The pitch gets harder.

But the burn math wins. At $14B a year, you don't get to keep products that aren't working.

The IPO either prices the potential or it doesn't close. At $840B, it's pricing potential so aggressively that any stumble — a quarterly miss, a competitor breakthrough, a regulation — hits hard.

Anthropic just had a model leak that called it a "step change." That's the kind of event that makes IPO bookbuilders nervous.

The math has to work before the window closes.