Amazon Web Services' annual tech conference AWS re:Invent has concluded its first official day of programming and has already delivered an countless stream of product news.
The unsurprising theme is AI for businesses, although this yr's focus is on upgrades that give customers more control in customizing AI agents – including one which AWS claims can learn from you after which work independently for days.
AWS re:Invent 2025, which runs through December 5, began with a keynote from AWS CEO Matt Garman, who discussed the concept AI agents can unlock the “true value” of AI.
“AI assistants are regularly giving strategy to AI agents that may perform and automate tasks in your behalf,” he said through the Dec. 2 keynote. “This is where we begin to see significant business returns out of your AI investments.”
While AI agent news is anticipated to be a everlasting presence throughout AWS re:Invent 2025, there have been other announcements as well. Here's a roundup of the announcements that caught our attention. TechCrunch will proceed to update this text until the top of AWS re:Invent. So remember to check back.
An AI training chip and Nvidia compatibility
AWS unveiled a new edition of its AI training chip called Trainium3 together with an AI system called UltraServer that runs it. The Tl;dr: This updated chip boasts some impressive specs, including the promise of as much as a fourfold increase in performance in each AI training and inference while reducing power consumption by 40%.
AWS also provided a teaser. The cloud provider already has Trainium4 in development, which might work with Nvidia's chips.
Techcrunch event
San Francisco
|
Thirteenth-Fifteenth October 2026
Advanced AgentCore features
AWS announced recent features in its AgentCore AI agent creation platform. A notable feature is the policy in AgentCore, which provides developers the power to more easily set limits for AI agents.
AWS also announced that agents will now give you the option to log and remember things about their users. Additionally, the corporate announced that it can help its customers rate agents using 13 pre-built rating systems.
A non-stop AI agent employee bee
AWS announced three recent AI agents (it's a term there again) called Frontier Agents, including one called the Kiro Autonomous Agent, which writes code and is designed to learn the way a team likes to work so that they can operate largely autonomously for hours or days.
Another of those recent agents handles security processes akin to code reviews, and the third handles DevOps tasks akin to incident prevention when recent code goes live. Preview versions of the agents can be found now.
New Nova models and services
AWS is introducing 4 recent AI models inside its Nova AI model family – three of that are text-generating and one which can generate text and pictures.
The company also announced a brand new service called Nova Forge that enables AWS cloud customers to access pre-trained, mid-training, or post-training models that they will then augment by training on their very own proprietary data. The big advantage of AWS is its flexibility and adaptableness.
Lyft's case for AI agents
The ride-hailing company was amongst many AWS customers spoke through the event to share their success stories and evidence of how products have impacted their business. Lyft is using Anthropic's Claude model through Amazon Bedrock to create an AI agent that handles questions and issues from drivers and riders.
The company said this AI agent reduced average resolution time by 87%. Lyft also said it has seen a 70% increase in driver use of its AI agent this yr.
An AI factory for the private data center
Amazon also announced “AI Factories,” allowing large corporations and governments to run AWS AI systems in their very own data centers.
The system was developed in collaboration with Nvidia and includes each Nvidia and AWS technology. While corporations using it could actually equip it with Nvidia GPUs, they also can go for Amazon's latest home-grown AI chip, the Trainium3. With the system, Amazon addresses data sovereignty, i.e. the necessity for governments and plenty of corporations to regulate their data and never share it, even in the case of using AI.
