The News
Anthropic’s Claude AI model is now running on 1 million of Amazon’s custom Trainium 2 AI processors, the company said Wednesday. Those processors are part of Amazon Web Services’ massive Project Rainier AI data center. Unlike most massive AI clusters, which are in one building or close together, Rainier is spread out across three states — Pennsylvania, Indiana, and Mississippi — according to a person familiar with the matter. The location in Pennsylvania has not previously been reported. The majority of chips are used for inference, and Anthropic uses a portion of the chips in evening hours for training runs. Those runs are paused during the day, when inference workloads increase.
In this article:
Know More
It’s difficult to compare Rainier with other massive compute clusters like xAI’s Rainier because they use different processors with varying strengths and weaknesses. Rainier may be the world’s most powerful AI cluster and an engineering feat either way, requiring roughly a gigawatt of power and generating a potential 1,300 exaflops (a measure of performance of a supercomputer’s speed) of compute, according to Amazon distinguished engineer Ron Diamant, who spoke with Semafor earlier this week.
Correction
Amazon’s massive data center is called Rainier. An earlier version of this article incorrectly referred to it as Colossus.