All the biggest news from AWS’ big tech show re:Invent 2025

The much-anticipated AWS re:Invent 2025 has commenced, showcasing a wealth of product announcements centered around artificial intelligence for enterprise applications. This year's focus is on empowering customers with enhanced tools to customize AI agents, including a groundbreaking agent that AWS claims can learn from user interactions and operate autonomously for extended periods. The event kicked off with an inspiring keynote from AWS CEO Matt Garman, who emphasized that AI agents are pivotal in unlocking the full potential of AI technology. "AI assistants are evolving into AI agents capable of executing tasks and automating processes on your behalf," he noted. "This shift is where businesses can start to see tangible returns on their AI investments." While AI agents are a central theme at this year's conference, several other notable announcements have also emerged. AWS introduced a new iteration of its AI training chip, dubbed Trainium3, alongside an AI infrastructure called UltraServer designed to leverage its capabilities. This upgraded chip boasts impressive specifications, promising up to a fourfold increase in performance for both AI training and inference, all while reducing energy consumption by 40%. Additionally, AWS hinted at the upcoming Trainium4, which will be compatible with Nvidia’s chips. New features have also been announced for the AgentCore platform, which facilitates the creation of AI agents. A particularly noteworthy addition is the Policy feature, allowing developers to easily define operational boundaries for their AI agents. Furthermore, these agents will now have the capability to log and retain information about their users, with AWS providing 13 pre-built systems to assist in evaluating these agents. Among the exciting new offerings are three AI agents labeled “Frontier agents.” One of these, the “Kiro autonomous agent,” is capable of writing code and is designed to adapt to a team's working style, enabling it to function independently for hours or even days. Another agent focuses on enhancing security processes, such as conducting code reviews, while the third is tasked with DevOps responsibilities, including preventing incidents during code deployments. AWS also expanded its Nova AI model family with four new AI models—three dedicated to text generation and one that can produce both text and images. A new service, Nova Forge, was introduced, enabling AWS customers to access models at various training stages, which they can further refine with their proprietary datasets. The core message from AWS is one of adaptability and customization. Many AWS customers, including Lyft, shared their success stories during the conference. Lyft has implemented Anthropic’s Claude model through Amazon Bedrock to develop an AI agent that addresses inquiries from drivers and riders, achieving an impressive 87% reduction in average resolution time and a 70% increase in driver engagement with the AI agent this year. Additionally, Amazon unveiled its “AI Factories” initiative, which allows large corporations and government entities to deploy AWS AI systems within their own data centers. Developed in collaboration with Nvidia, this system incorporates both Nvidia technology and AWS components. Organizations using this solution can choose between Nvidia GPUs or Amazon's latest Trainium3 chip, addressing the critical issue of data sovereignty by allowing entities to maintain control over their data while utilizing AI capabilities.