The new seven-year agreement doesn’t put a gigawatt number on the compute OpenAI is buying, but it costs roughly $40B to build out a 1 GW data center.
Massive compute needs
— Scaling frontier AI requires massive, reliable compute, said OpenAI co-founder and CEO Sam Altman on their announcement page. — Our partnership with AWS strengthens the broad compute ecosystem that will power this next era and bring advanced AI to everyone.
The deal would provide «immediate» access to AWS’ cluster of some 800,000 Nvidia GPUs (although this number has surely risen), along with Amazon’s own, custom AI processors.
The capacity is «targeted to be deployed» in 2026, and can also be expanded into 2027 «and beyond.»
The compute power will be used for both inference in day-to-day operations for ChatGPT, and in terms of training new models, OpenAI says.
Tall investments for OpenAI
OpenAI is now on tap to invest well over $1.4 trillion in infrastructure deals, and questions remain over their ability to follow through.
Asked lately about how they plan to achieve that on $13 billion of annualized revenue, Sam Altman said on the Bg2Pod that:
— We’re doing well more revenue than that, and suggested they might hit $100 billion in revenue as early as 2027.
— Revenue is growing steeply, said Altman. — We are taking a forward bet that it’s going to continue to grow and that not only will ChatGPT keep growing, but we will be able to become one of the important AI clouds.
He then added that their consumer device would be a major business driver, and that AI’s promise to automate science will create «huge value.»
Read more: OpenAI’s announcement, writeups on Reuters, AP and Engadget. Discussion on r/Singularity.