Amazon Web Services’ annual tech conference, AWS re:Invent, has concluded, with a central focus on AI for enterprises.
This year emphasized upgrades that provide customers more control to personalize AI agents. AWS claims one of these agents can learn from users and operate independently for days. Amazon CTO Dr. Werner Vogels wrapped up the conference with a keynote that aimed to reassure developers regarding concerns about AI potentially taking over engineering roles.
AWS re:Invent 2025, which runs until December 5, kicked off with a keynote from AWS CEO Matt Garman, who highlighted that AI agents can unlock the “true value” of AI.
“AI assistants are giving way to AI agents that can take action and automate tasks on your behalf,” he said on December 2. “This is where we begin to see real business returns from your AI investments.”
On December 3, the AI agent theme continued, along with deeper dives into customer experiences. Swami Sivasubramanian, vice president of Agentic AI at AWS, gave one of the keynote addresses, expressing a very optimistic outlook. “We are living in times of great change,” he said. “For the first time, we can describe goals in natural language, and agents create the plan. They write the code, recommend necessary tools, and execute the entire solution. Agents provide you the freedom to build without limits, speeding up the journey from idea to impactful results.”
While AI agent discussions dominated AWS re:Invent 2025, several other announcements stood out. TechCrunch will update this article with the latest insights at the top throughout the conference, so be sure to check back.
Werner out …
Amazon CTO Werner Vogels delivered the conference’s closing keynote, marking what he indicated would be his final appearance.
“This is my last re:Invent keynote,” he said, quickly clarifying he’s not leaving the company. “I’m not leaving Amazon or anything like that, but after 14 re:Invents, you guys deserve young, fresh voices.”
Vogels engaged with a packed audience for over an hour, concluding with “Werner, out” and a mic drop.
Will AI take your job?
Vogels addressed the topic of AI and its future impact, including concerns about job displacement.
“Will AI take my job? Maybe,” he pondered, then explained that while some tasks will be automated, not all skills will become irrelevant. “So maybe we should rethink this question. Will AI make me obsolete? Absolutely not, if you evolve.”
Next-gen CPU
On Thursday, AWS introduced its Graviton5 CPU, touted to be the highest performing and most efficient chip yet. The Graviton5 features 192 processor cores and a design that minimizes data travel between cores, reducing inter-core communication latency by up to 33% while increasing bandwidth, according to the company.
Doubling down on LLMs
AWS unveiled new tools for businesses to create their own models. Specifically, AWS is enhancing Amazon Bedrock and Amazon SageMaker AI to facilitate building custom large language models (LLMs). For instance, SageMaker will now offer serverless model customization, allowing developers to start model building without worrying about compute resources or infrastructure. This feature can be accessed via either a self-directed path or by prompting an AI agent. AWS also launched Reinforcement Fine Tuning in Bedrock, enabling developers to select a preset workflow or reward system and have Bedrock execute their customization process autonomously.
Andy Jassy shares some numbers
Amazon CEO Andy Jassy took to the social media platform X to elaborate on Matt Garman’s keynote, emphasizing that the current generation of its AI chip, Trainium2, is already generating significant revenue.
His remarks coincided with the introduction of the next-generation chip, Trainium3, hinting at a bright financial future for the product.
Database savings arrives
Among the numerous announcements, one detail has already been celebrated: discounts.
AWS has announced Database Savings Plans, allowing customers to reduce database costs by up to 35% when they commit to a consistent usage rate over a one-year term. The savings will automatically apply every hour to eligible usage across supported database services, with any additional usage beyond the commitment billed at on-demand rates.
Corey Quinn, chief cloud economist at Duckbill, summed it up perfectly in his blog: “Six years of complaining finally pays off.”
Can’t get a better deal than free, Amazon hopes
Can another AI coding tool capture the interest of startup founders? Amazon believes that offering a year’s worth of credits for free will do the trick for its product, Kiro. The company will provide credits to Kiro Pro+ for qualified startups that apply for the deal by the end of the month, although only early-stage startups in select countries are eligible.
An AI training chip and Nvidia compatibility
AWS unveiled an upgraded version of its AI training chip, Trainium3, along with a new AI system called UltraServer. This enhanced chip boasts impressive specs, promising up to 4x performance gains for both AI training and inference while reducing energy consumption by 40%.
AWS also teased that Trainium4 is in development, which will be compatible with Nvidia’s chips.
Expanded AgentCore capabilities
AWS announced new features for its AgentCore AI agent building platform. Notably, the Policy feature allows developers to set boundaries more easily for their AI agents.
Additionally, agents can now log and remember details about their users. AWS also introduced 13 prebuilt evaluation systems to help customers assess their agents.
A nonstop AI agent worker bee
AWS announced three new AI agents, collectively termed “Frontier agents,” including “Kiro autonomous agent,” which is designed to write code and learn how a team prefers to work, allowing it to largely operate independently for hours or days.
Another agent focuses on security procedures like code reviews, while the third performs DevOps tasks aimed at preventing incidents when deploying new code. Preview versions of these agents are already available.
New Nova models and services
AWS is launching four new AI models in its Nova model family—three focused on text generation and one that can create both text and images.
Additionally, AWS introduced a new service called Nova Forge, allowing customers to access pre-trained, mid-trained, or post-trained models that they can further refine using their proprietary data. AWS’s primary pitch is on flexibility and customizability.
Lyft’s argument for AI agents
Ride-hailing company Lyft shared its success story during the event, highlighting how it uses Anthropic’s Claude model through Amazon Bedrock to create an AI agent that addresses driver and rider inquiries.
Lyft reported an 87% reduction in average resolution time and a 70% increase in driver engagement with the AI agent this year.
An AI Factory for the private data center
Amazon also debuted “AI Factories,” enabling major corporations and governments to run AWS AI systems within their own data centers.
Developed in collaboration with Nvidia, the system integrates both Nvidia’s technology and AWS’s. Companies can either equip it with Nvidia GPUs or opt for Amazon’s latest AI chip, Trainium3. This initiative addresses data sovereignty concerns, allowing organizations to control their data without sharing it for AI use.
Check out the latest announcements covering everything from agentic AI and cloud infrastructure to security and more from AWS’s flagship event in Las Vegas. This video is presented in partnership with AWS.
