Amazon Web Services (AWS) announced plans for an AI supercomputer named 'Ultracluster' on December 5, 2024. This project, part of 'Project Rainier,' involves hundreds of thousands of in-house Trainium chips and aims to be one of the largest supercomputers globally for training AI models.
According to AWS Vice President Dave Brown, 'Ultracluster' is set for completion in 2025 and will serve AI startup Anthropic, which recently secured an additional $4 billion funding from AWS, raising Amazon's total investment to $8 billion.
Anthropic, founded by former OpenAI executives, is recognized as a competitor to OpenAI and has developed the Claude generative AI software. AWS customers can now fine-tune their data using Claude's features.
At the annual re:Invent conference in Las Vegas, AWS also introduced a new server called 'Ultraserver,' comprising 64 interconnected chips. The announcements underscore AWS's strategy to promote its custom-designed Trainium chips as a competitive alternative to Nvidia's GPUs.
AWS CEO Matt Garman noted the current dominance of Nvidia in the GPU market and expressed the company's intent to provide customers with more choices. Amazon CEO Andy Jassy emphasized that Trainium chips offer 30% to 40% better price performance than existing GPU instances, describing them as a 'game changer.'