Amazon Shifts AI Chip Strategy, Focuses on Trainium to Compete with Nvidia

AWS discontinues Inferentia chip development, expands Trainium capabilities, and announces next-gen Trainium3 for 2025.

Amazon Web Services (AWS) has halted development of its Inferentia AI chip to concentrate on its Trainium chip line for both training and inference tasks.
AWS unveiled the Trainium2 processor, which powers new EC2 Trn2 instances and UltraServers, offering significant performance improvements for large AI workloads.
The company announced plans for Trainium3, expected in 2025, built on TSMC's 3nm process and projected to deliver four times the performance of Trainium2 with a 40% efficiency boost.
AWS is constructing an ExaFLOPS-class supercomputer using hundreds of thousands of Trainium2 processors, targeting advanced AI model training and competing with Nvidia's dominance in the AI chip market.
Customers such as Apple and AI startup Anthropic are already adopting Trainium2, while AWS plans to introduce a new AI model, codenamed 'Olympus,' potentially leveraging Trainium3 capabilities.