All the biggest news from AWS’ big tech show re:Invent 2025

Amazon Net Providers’ annual tech convention AWS re:Invent has wrapped. And the singular message, amid a deluge of product information and keynotes, was AI for the enterprise.

This yr it was all about upgrades that give clients larger management to customise AI brokers, together with one which AWS claims can be taught from you after which work independently for days. Amazon CTO Dr. Werner Vogels capped off the ultimate evening with a keynote geared toward lifting up builders and assuaging any fears that AI is coming for engineering jobs.

AWS re:Invent 2025, which runs via December 5, began with a keynote from AWS CEO Matt Garman, who leaned into the concept that AI brokers can unlock the “true worth” of AI.

“AI assistants are beginning to give strategy to AI brokers that may carry out duties and automate in your behalf,” he stated through the December 2 keynote. “That is the place we’re beginning to see materials enterprise returns out of your AI investments.”

On December 3, the convention pressed on with its AI brokers messaging, in addition to deeper dives into buyer tales. Swami Sivasubramanian, vice chairman of Agentic AI at AWS, gave one of many keynote talks. To say he was bullish is probably understating the vibe.

“We live in instances of nice change,” Sivasubramanian stated through the discuss. “For the primary time in historical past, we are able to describe what we need to accomplish in pure language, and brokers generate the plan. They write the code, name the mandatory instruments, and execute the entire answer. Brokers provide the freedom to construct with out limits, accelerating how shortly you’ll be able to go from concept to influence in an enormous approach.”

Whereas AI agent information guarantees to be a persistent presence all through AWS re:Invent 2025, there have been different bulletins, too. Here’s a roundup of those that received our consideration. TechCrunch will replace this text, with the latest insights on the high, via the top of AWS re:Invent. Be sure you examine again.

Werner out …

Amazon CTO Werner Vogels had the closing keynote of the convention — and it seems like this might be his final one.

“That is my closing re:Invent keynote,” he stated, then shortly added he’s not leaving the corporate. “I’m not leaving Amazon or something like that, however I feel that after 14 re:Invents you guys are owed younger, recent, new voices.”

Vogels then spent greater than an hour speaking to a packed room earlier than ending with a “Werner, out” and a literal mic drop.

Will AI take your job?

Vogels spent a lot of the closing keynote speaking about AI and its function sooner or later, together with the looming menace that it’s going to take away jobs.

“Will AI take my job? Perhaps,” Vogels requested and answered, earlier than noting that some duties might be automated, and a few expertise will turn out to be out of date. “So possibly we should always rephrase and reframe this query. Will AI make me out of date? Completely not, if you happen to evolve.”

Subsequent-gen CPU

AWS unveiled its Graviton5 CPU on Thursday, the next-generation chip that the corporate guarantees might be its highest performing, most effective but. The Graviton5 comprises 192 processor cores, a dense and environment friendly design that AWS says reduces the gap knowledge should journey between cores. That helps minimize inter-core communication latency by as much as 33% whereas growing bandwidth, the corporate stated. 

Doubling down on LLMs

AWS introduced extra instruments for enterprise clients to create their own models. Particularly, AWS stated it’s including new capabilities for each Amazon Bedrock and Amazon SageMaker AI to make constructing customized LLMs simpler.

For example, AWS is bringing serverless mannequin customization to SageMaker, which permits builders to start out constructing a mannequin while not having to consider compute assets or infrastructure. The serverless mannequin customization may be accessed via both a self-guided path or by prompting an AI agent.

AWS additionally introduced Reinforcement Advantageous Tuning in Bedrock, which permits builders to decide on a preset workflow or reward system and have Bedrock run their customization course of mechanically from begin to end.

Andy Jassy shares some numbers

Amazon CEO Andy Jassy took to social media platform X to expound on AWS chief Matt Garman’s keynote speech. The message: The present technology of its Nvidia-competitor AI chip Trainium2 is already bringing in loads of cash.

His feedback had been tied to the reveal of its next-generation chip, Trainium3, and meant to forecast a promising income future for the product.

Database financial savings arrives

Tucked among the many dozens of bulletins is one merchandise that’s already getting cheers: Reductions.

Particularly, AWS stated it was launching Database Financial savings Plans, which assist clients cut back database prices by as much as 35% after they decide to a constant quantity of utilization ($/hour) over a one-year time period. The corporate stated the financial savings will mechanically apply every hour to eligible utilization throughout supported database providers, and any extra utilization past the dedication is billed at on-demand charges.

Corey Quinn, chief cloud economist at Duckbill, summed it up properly in his blog post, “Six years of complaining lastly pays off.”

Can’t get a greater deal than free, Amazon hopes

Is there any approach for one more AI coding instrument to win the hearts of startup founders? Amazon hopes a yr’s value of credit, without cost, will do the trick for its providing, Kiro. The company will be giving away credits to Kiro Pro+ to certified startups that apply for the deal earlier than the top of the month. Nevertheless, solely early-stage startups in sure nations are eligible.

An AI coaching chip and Nvidia compatibility

AWS launched a new version of its AI training chip known as Trainium3 together with an AI system known as UltraServer that runs it. The TL;DR: This upgraded chip comes with some spectacular specs, together with a promise of as much as 4x efficiency positive aspects for each AI coaching and inference whereas reducing power use by 40%.

AWS additionally offered a teaser. The corporate already has Trainium4 in improvement, which is able to be capable to work with Nvidia’s chips.

Expanded AgentCore capabilities

AWS introduced new features in its AgentCore AI agent building platform. One function of notice is Coverage in AgentCore, which provides builders the power to extra simply set boundaries for AI brokers.

AWS additionally introduced that brokers will now be capable to log and keep in mind issues about their customers. Plus it introduced that it’s going to assist its clients consider brokers via 13 prebuilt analysis programs.

A nonstop AI agent employee bee

AWS introduced three new AI brokers (there’s that time period once more) known as “Frontier brokers,” together with one known as “Kiro autonomous agent” that writes code and is designed to find out how a crew likes to work so it can operate largely on its own for hours or days.

One other of those new brokers handles safety processes like code opinions, and the third does DevOps duties such as stopping incidents when pushing new code reside. Preview variations of the brokers can be found now.

New Nova fashions and providers

AWS is rolling out four new AI models inside its Nova AI mannequin household — three of that are textual content producing and one that may create textual content and pictures.

The corporate additionally introduced a brand new service known as Nova Forge that permits AWS cloud clients to entry pre-trained, mid-trained, or post-trained fashions that they’ll then high off by coaching on their very own proprietary knowledge. AWS’s huge pitch is flexibility and customization.

Lyft’s argument for AI brokers

The ride-hailing firm was amongst many AWS clients that piped up during the event to share their success tales and proof of how merchandise affected their enterprise. Lyft is utilizing Anthropic’s Claude mannequin through Amazon Bedrock to create an AI agent that handles driver and rider questions and points.

The corporate stated this AI agent has diminished common decision time by 87%. Lyft additionally stated it has seen a 70% enhance in driver utilization of the AI agent this yr.

An AI Manufacturing facility for the non-public knowledge middle

Amazon additionally announced “AI Factories” that permit huge firms and governments to run AWS AI programs in their very own knowledge facilities.

The system was designed in partnership with Nvidia and contains each Nvidia’s tech and AWS’s. Whereas corporations that use it might probably inventory it with Nvidia GPUs, they’ll additionally go for Amazon’s latest homegrown AI chip, the Trainium3. The system is Amazon’s approach of addressing knowledge sovereignty, or the necessity of governments and lots of corporations to manage their knowledge and never share it, even to make use of AI.

Try the most recent reveals on all the pieces from agentic AI and cloud infrastructure to safety and rather more from the flagship Amazon Net Providers occasion in Las Vegas. This video is delivered to you in partnership with AWS.