OpenAI just announced annualized revenue crossed $20 billion - up from $6 billion in 2024. That's spectacular growth. Tripling revenue in a year is founder-fantasy territory.

Now let's talk about what they're not announcing: whether any of that revenue translates to actual profit.

Spoiler: it doesn't.

Reports suggest OpenAI is burning through billions in compute costs, with losses projected to hit $14 billion in 2026. If those numbers hold, the company could face a cash shortfall by 2027 despite the revenue growth.

This is the unit economics problem every AI company is desperately trying to ignore.

Here's the math: training and running large language models costs enormous amounts of money. GPT-5's training run reportedly cost hundreds of millions. Inference - actually answering user queries - costs real money per token. At scale, those costs are staggering.

OpenAI charges $20/month for ChatGPT Plus. Enterprise contracts cost more, but even at enterprise pricing, you're competing against compute costs that don't compress easily.

The bet is that scale brings efficiency. Train bigger models on more data, optimize inference, build custom chips, and eventually the economics work. Maybe they're right. Amazon lost money for years before AWS turned it into a profit machine.

But Amazon was losing money on logistics and warehousing - capital investments that became assets. OpenAI is losing money on compute that evaporates the moment you use it.

Every query burns money. Every training run burns more. And unlike AWS, where marginal cost per customer approaches zero, each OpenAI user actively costs resources.

The revenue growth is real. 20 billion dollars is real money. But it's not clear that growth path leads to profitability at current cost structures.

What happens if it doesn't? OpenAI has raised billions from Microsoft and others. They can keep raising. But eventually, investors want returns, not just revenue graphs going up and to the right.

The entire AI industry is playing this game. Anthropic, Google's DeepMind, even 's AI division - they're all spending billions to build models that may or may not ever be profitable.