Aptly named, Amazon Trainium chips are used to train AI models on Amazon's EC2 cloud computing service using Trn2 server instances. In 2025, Trainium3 chips are expected to be more energy efficient and twice as fast as Trainium2.
Amazon Inferentia chips, also appropriately named, are used in Inf2 server instances for system execution, which is known as "inference." For the highest performance training and inference, Amazon offers the EC2 UltraServers and UltraClusters. UltraServer clusters comprise thousands of Trainium chips with considerably more memory and bandwidth. See
Amazon Bedrock and
EC2.