Core Insights - OpenAI has committed to utilizing approximately 2 GW of Trainium computing capacity through AWS infrastructure to support stateful runtime, Frontier, and other advanced workload requirements [2] Group 1 - The commitment includes the use of AWS's released Trainium3 and the next-generation Trainium4 chips [2] - Trainium4 is expected to begin delivery in 2027, featuring significantly enhanced FP4 computing performance, wider memory bandwidth, and higher video memory capacity [2]