tech
Internal Amazon documents reveal a major overhaul of one of the world's biggest AI data centers
Amazon is redesigning part of its Indiana AI campus into a massive Trainium-powered cluster for its next frontier AI model, internal documents show.
TL;DR
- Amazon is reconfiguring a large AI data center campus in Indiana to build its next frontier AI models.
- The project involves consolidating multiple data centers and deploying thousands of Trainium-powered AI servers into an 'AGI SuperCluster'.
- This effort, part of a broader 'AGI Pivot' initiative, aims to accelerate the training of future in-house AI models.
- The redesign seeks to maximize returns on Amazon's AI infrastructure investments by improving efficiency and utilizing existing facilities.
- The Indiana campus also houses facilities for Anthropic's AI supercomputer, Project Rainier, but the new initiative will not affect those servers.
- Amazon is accelerating deployment schedules, with an 'emergent request' for over 6,000 Trainium servers to prepare a new AI model for the re:Invent conference.
- The company is increasing its projected 2026 capital expenditures to $220 billion, reflecting strong demand for AI computing.
- Amazon's custom chip business, including Trainium AI chips, is on track to generate over $25 billion in annual revenue.
- The new models require expanded data storage to support multimodal AI training, involving high-resolution images.
- The push for advanced AI development continues with urgency, despite recent layoffs in Amazon's AGI organization.