AWS re:Invent 2025, Amazon Web Services' annual flagship tech conference, recently concluded another day in Las Vegas, delivering a torrent of announcements centered on the future of enterprise AI. From advanced AI agents and powerful new AI chips to enhanced LLM customization and significant database savings, the event underscored AWS's commitment to empowering businesses with cutting-edge cloud solutions. The overarching theme revolved around giving customers greater control and flexibility in leveraging artificial intelligence.

The conference, which runs through December 5, kicked off with a keynote from AWS CEO Matt Garman, who championed the idea that AI agents are key to unlocking AI's "true value." Garman highlighted the shift from AI assistants to more autonomous AI agents capable of performing complex tasks and automating processes on behalf of users, predicting "material business returns" from these investments.

On December 3, Swami Sivasubramanian, Vice President of Agentic AI at AWS, echoed this sentiment in his keynote. He emphasized the transformative potential of agents, stating, "We are living in times of great change. For the first time in history, we can describe what we want to accomplish in natural language, and agents generate the plan. They write the code, call the necessary tools and execute the complete solution. Agents give you the freedom to build without limits, accelerating how quickly you can go from idea to impact in a big way."

While AI agents dominated the conversation, several other significant announcements captured attention.

Doubling Down on Custom LLMs

AWS introduced new tools designed to simplify the creation of custom large language models (LLMs) for enterprise customers. Enhancements were rolled out for both Amazon Bedrock and Amazon SageMaker AI.

For SageMaker, AWS is bringing serverless model customization, allowing developers to build models without needing to manage compute resources or infrastructure. This feature can be accessed through either a self-guided path or by prompting an AI agent. Additionally, Bedrock now features Reinforcement Fine Tuning, enabling developers to automate their customization process from start to finish by selecting a pre-set workflow or reward system.

Amazon CEO Andy Jassy Shares Chip Success

Amazon CEO Andy Jassy took to social media platform X to elaborate on AWS chief Matt Garman's keynote, revealing that the current generation of AWS's Nvidia-competitor AI chip, Trainium2, is already a multi-billion dollar business. His comments were strategically timed with the unveiling of the next-generation chip, Trainium3, signaling a promising revenue future for the product.

Database Savings Plans Arrive

Among the numerous announcements, one item garnered particular enthusiasm: new discounts. AWS launched Database Savings Plans, offering customers up to a 35% reduction in database costs. This saving applies when customers commit to a consistent amount of usage ($/hour) over a one-year term, automatically applying to eligible usage across supported database services. Any usage exceeding the commitment is billed at standard on-demand rates. Corey Quinn, chief cloud economist at Duckbill, humorously summarized the sentiment in his blog post titled, "Six years of complaining finally pays off."

Next-Gen AI Training Chip and Nvidia Compatibility

AWS unveiled a new version of its AI training chip, Trainium3, alongside an accompanying AI system called UltraServer. This upgraded chip boasts impressive specifications, promising up to 4x performance gains for both AI training and inference, while simultaneously reducing energy consumption by 40%. Looking ahead, AWS also teased the development of Trainium4, which will be designed to work seamlessly with Nvidia's chips, indicating a strategic move towards broader compatibility.

Expanded AgentCore Capabilities and New Frontier Agents

Further solidifying its focus on AI agents, AWS announced new features for its AgentCore AI agent building platform. A notable addition is 'Policy in AgentCore,' which empowers developers to more easily set boundaries and guardrails for AI agents. AWS also revealed that agents will now be able to log and remember user interactions, and customers will benefit from 13 prebuilt evaluation systems to assess agent performance.

The company also previewed three new "Frontier agents." Among them is the "Kiro autonomous agent," designed to write code and learn team workflows, enabling it to operate largely independently for extended periods. The other new agents specialize in security processes, such as code reviews, and DevOps tasks like preventing incidents during new code deployments. Preview versions of these agents are currently available.

Introducing New Nova Models and Services

AWS is expanding its Nova AI model family with the rollout of four new models. Three of these are dedicated to text generation, while one offers both text and image creation capabilities.

Additionally, AWS announced Nova Forge, a new service that provides AWS cloud customers access to pre-trained, mid-trained, or post-trained models. Customers can then further customize these models by training them on their own proprietary data, emphasizing AWS's commitment to flexibility and customization.

Lyft's Argument for AI Agents

Ride-hailing giant Lyft was among several AWS customers who shared their success stories, demonstrating the tangible business impact of AWS products. Lyft is leveraging Anthropic's Claude model through Amazon Bedrock to power an AI agent that handles driver and rider inquiries. This implementation has reportedly reduced average resolution time by an impressive 87% and seen a 70% increase in driver usage of the AI agent this year.

AI Factories for Private Data Centers

In a move to address data sovereignty concerns, Amazon also announced "AI Factories." These allow large corporations and governments to run AWS AI systems within their own private data centers. Developed in partnership with Nvidia, the system integrates both Nvidia's technology and AWS's. While users can equip these factories with Nvidia GPUs, they also have the option to utilize Amazon's latest homegrown AI chip, the Trainium3. This initiative provides a solution for organizations requiring strict control over their data, even when utilizing advanced AI capabilities.

Check out the latest reveals on everything from agentic AI and cloud infrastructure to security and much more from the flagship Amazon Web Services event in Las Vegas. This video is brought to you in partnership with AWS.