At AWS re:Invent, Amazon Web Services (AWS) announced the general availability of EC2 Trn3 UltraServers, powered by the new Trainium3 chip—the first AI chip built on a 3nm process. These servers are designed to help organizations of all sizes train and deploy large-scale AI models faster and at significantly lower costs. Trainium3 UltraServers deliver up to 4.4 times more compute performance, 4 times greater energy efficiency, and nearly 4 times higher memory bandwidth compared to their predecessors, the Trainium2 UltraServers. This leap in performance enables faster AI development cycles while reducing operational expenses. The new servers can scale up to 144 Trainium3 chips, delivering up to 362 FP8 PFLOPs of compute power and offering 4 times lower latency. This makes them ideal for training massive AI models quickly and serving inference workloads at scale with high responsiveness. Early adopters are already seeing tangible benefits. Anthropic, Karakuri, Metagenomi, NetoAI, Ricoh, and Splash Music are reducing both training and inference costs by up to 50% using Trainium technology. Decart has achieved four times faster inference for real-time generative video at just half the cost of traditional GPU-based solutions. Additionally, Amazon Bedrock is now running production workloads on Trainium3, demonstrating its readiness for mission-critical AI applications. AWS continues to expand its AI infrastructure leadership with a focus on performance, efficiency, and accessibility. The Trainium3 chip represents a major milestone in AWS’s strategy to provide customers with powerful, cost-effective tools to accelerate innovation in generative AI and beyond. With one of the most comprehensive AI service portfolios and the broadest global infrastructure, AWS empowers developers, enterprises, and innovators to bring ambitious ideas to life. For more information, visit aws.amazon.com and follow @AWSNewsroom.