AWS Launches Trainium2 AI Chips and Services

Amazon Web Services (AWS) has launched its new Trainium2 AI chips, which are now available through EC2 Trn2 instances and Trn2 UltraServers, designed for training and deploying large language models (LLMs). The Trn2 instances, equipped with 16 Trainium2 chips, offer up to 20.8 peak petaflops of compute, while the UltraServers, featuring 64 interconnected chips, can scale up to 83.2 petaflops. These new offerings provide a significant 30-40% improvement in price performance compared to current GPU-based EC2 instances. AWS is collaborating with Anthropic on Project Rainier to create a massive cluster of Trn2 UltraServers, enhancing their AI model training capabilities. Additionally, AWS has announced plans for the next-generation Trainium3 chips, expected in late 2025, which will be twice as performant and more energy efficient. This move positions AWS as a competitive alternative to Nvidia's GPUs in the growing AI market.