The annual Tech Conference Amazon Web Re: Invent has wrapped up the official business day and has delivered endless product news.
An incongruous theme is AI for the company, although this year it’s all about upgrades that give customers control over managing AWS agents that can learn from you and then work autonomously.
AWS RE: INVENT 2025which ran through December 5, starting with a keynote from the CEO of AWS that inhibited the idea that AI agents could reveal “true value”.
“AI assistants are starting to give you an AI way to do your tasks and automate them,” he said in a December 2 note. “This is where we start to see a material business return on your ai investment.”
While news agent AI promises to be a constant presence during AWS Re:Invent 2025, there are other announcements as well. Here’s a roundup of the announcements that got attention. TechCrunch will continue to update this article through the end of alws Re:Invent so be sure to check back.
Ai training chip and nvidia compatibility
AWS introduced A A new version of the AI training chip It’s called Trainium3 along with an AI system called ultraserver running on it. TL; Dr: this upgraded chip comes with some impressive specs, including promising up to 4x performance for AI training and inference while reducing energy consumption by 40%.
AWS also provided a teaser. Cloud providers already have Training Trainum4, which will work with Nvidia chips.
TechCrunch events
San Francisco
I’m fat
October 13-15, 2026
Expanded Agentganding Capabilities
AWS announced New features in Agent Agent building platformSee rank-. One feature of note is the policy in Agent, which gives developers the ability to set limits for Ai agents.
AWS also announced agents will now be able to log and remember users. It added that it will help customers evaluate agents through 13 built-in evaluation systems.
The unstoppable Aven bee
AWS announced three new Ai agents (there’s that term again) called “frontier agents” including the “Kodo Autonomous Agent” that is said to learn how teams like to work So it can be used extensively for hours or days.
Another new agent handles security processes like code review, and a third performs tasks like preventing incidents when pushing new code. A preview version of the agent is available now.
New Nova Models and Services
AWS ROLLING OUT four new AI models In the Model Nova AI model – three of which are text generators and one that can create text and images.
The company also announced a new service called Nova Forge that allows pre-trained, mid-satisfied or post-trained cloud models to be created by training on their own property data. The big thing is flexibility and customization.
Lyft’s goal for Agent Ai
Ride-Hailing companies have a large number of AWS piped during the event to share success stories and evidence of products that have impacted business. Lyft uses the anthropic Claude model through Amazon Bedrock to create an AI Agent that handles driver and rider issues.
The company says this AI Agent has reduced the average resolution by 87%. Lyft also says it has seen a 70% increase in the use of its AI Agent drivers this year.
AI factories for private data centers
Amazon too announce the “AI factory” This allows large companies and governments to run AI systems in their own data centers.
The system was designed in partnership with Nvidia, and includes Nvidia technology and AWs. While companies that use it can provide GPUs on NVIDIA, they can also choose the latest Homegrown Homombrown chip on Amazon, the Trainum3. The system is a way to deal with data tyranny, or the need for governments and many companies to control data and not share it, even with the use of AI.