OpenAI commits $750 billion to compute through 2030

OpenAI has committed roughly $750 billion to compute, data centers, and chips through 2030, per a TechCrunch tally published July 22. That's close to Sweden's entire GDP, signed by a company whose current revenue is a fraction of it.

The number isn't one contract. It's TechCrunch's sum of separate deals for compute, data center builds, and chip purchases. And it's commitments, not cash OpenAI already has, which makes the funding question sharper, not softer. Nobody signs three-quarters of a trillion dollars in obligations against revenue they're already collecting. They sign it against revenue they're betting on.

What builders should take from this

This isn't OpenAI betting models get smarter. It's OpenAI betting inference capacity is scarce enough to justify pre-committing that much money to secure it. I read that as a signal about capacity, not a headline number: the constraint the labs are pricing against is capacity, not quality.

If that's right, the assumption a lot of us build on, that inference keeps getting cheaper per unit of capability, stops holding at some point. Plan your unit economics for a world where capability costs more once demand catches up to what labs can actually serve, not less. I'd rather size margins for that world now than get surprised by it later.

TechCrunch ran this tally on July 22, and the total is still moving as more of these deals get signed.