Sam Altman is sounding the alarm: the world is about to run out of compute. As large AI models get hungrier by the month, OpenAI is launching a "capacity guarantee" program that lets enterprises lock in GPU access for one to three years — and get tiered token discounts based on how much they're willing to spend.
https://twitter.com/sama/status/2056827105401614656
It's a straightforward trade. Companies get predictable compute for their AI agents and production systems; OpenAI gets upfront cash to plan and build the next generation of data centers. Altman says this is essentially a long-term hedge against what he sees as a permanent compute crunch.
But there's a hard cap. Altman made clear that once the allocated capacity is sold out, that's it. And OpenAI will always reserve enough compute to keep its own core products — ChatGPT, Codex, and the like — running smoothly.