All the major highlights from AWS’s flagship technology event re:Invent 2025

AWS re:Invent 2025: Major Announcements and Highlights

The first official day of AWS re:Invent 2025 has concluded, bringing with it a flood of new product announcements from Amazon Web Services’ annual technology conference.

This year, the spotlight is firmly on enterprise AI, with a particular emphasis on empowering customers to tailor AI agents to their needs. Notably, AWS introduced an agent that can learn from user behavior and operate autonomously for extended periods.

The event, which runs until December 5, kicked off with a keynote by AWS CEO Matt Garman. He underscored the transformative potential of AI agents, stating, “AI assistants are evolving into AI agents capable of executing tasks and automating processes on your behalf. This is where organizations are beginning to see tangible returns on their AI investments.”

While AI agents are a recurring theme at this year’s conference, AWS also unveiled a range of other innovations. Below is a summary of the most noteworthy announcements, with ongoing updates expected throughout the event.

Introducing Trainium3 and Enhanced Nvidia Integration

AWS revealed the latest iteration of its AI training chip, Trainium3, alongside a new AI system called UltraServer. This upgraded chip promises up to four times the performance for both AI training and inference, while reducing energy consumption by 40%.

Looking ahead, AWS teased the development of Trainium4, which will be compatible with Nvidia’s hardware.

AgentCore Platform Gets New Features

The AgentCore platform, AWS’s toolkit for building AI agents, received several enhancements. Among them is the new Policy feature, enabling developers to more easily define operational boundaries for their AI agents.

Additional updates include the ability for agents to record and recall user-specific information, as well as access to 13 prebuilt evaluation systems to help customers assess agent performance.

Frontier Agents: Autonomous AI for the Enterprise

AWS introduced three new “Frontier agents,” including the Kiro autonomous agent, which is designed to learn a team’s workflow and independently write code for hours or even days at a time.

Another agent focuses on automating security tasks such as code reviews, while the third handles DevOps responsibilities like incident prevention during code deployment. Preview versions of these agents are now available.

New Nova AI Models and Nova Forge Service

Four new models have been added to AWS’s Nova AI family—three specializing in text generation and one capable of producing both text and images.

AWS also launched Nova Forge, a service that lets customers access models at various training stages—pre-trained, mid-trained, or post-trained—and further refine them using their own proprietary data. The emphasis is on flexibility and customization for enterprise users.

Lyft Shares Success with AI Agents

During the event, several AWS customers shared their experiences, including ride-hailing company Lyft. Leveraging Anthropic’s Claude model through Amazon Bedrock, Lyft developed an AI agent to handle driver and rider inquiries.

According to Lyft, this AI agent has cut average resolution times by 87% and led to a 70% increase in driver engagement with the system over the past year.

AI Factory: Bringing AWS AI to Private Data Centers

Amazon also introduced “AI Factories,” enabling large organizations and government entities to deploy AWS AI solutions within their own data centers.

Developed in collaboration with Nvidia, these systems incorporate both Nvidia and AWS technologies. Customers can choose between Nvidia GPUs or Amazon’s new Trainium3 chip, addressing the need for data sovereignty by allowing sensitive data to remain on-premises while utilizing advanced AI capabilities.

Sponsored: Watch Industry Streams from re:Invent

Explore the latest developments in agentic AI, cloud infrastructure, security, and more from AWS’s flagship event in Las Vegas. This video content is presented in partnership with AWS.