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

Amazon Web Services’ yearly tech league AWS re:Invent has wrapped up its archetypal authoritative time of programming and has already delivered an endless watercourse of merchandise news.

The unsurprising taxable is AI for the enterprise, though this twelvemonth it’s each astir upgrades that springiness its customers greater power to customize AI agents — including 1 that AWS claims tin larn from you and past enactment independently for days.

AWS re:Invent 2025, which runs done Dec. 5, started with a keynote from AWS CEO Matt Garman who leaned into the thought that AI agents tin unlock the “true value” of AI.

“AI assistants are starting to springiness mode to AI agents that tin execute tasks and automate connected your behalf,” helium said during the December 2 keynote. “This is wherever we’re starting to spot worldly concern returns from your AI investments.”

While AI cause quality promises to beryllium a persistent beingness passim AWS re:Invent 2025, determination were different announcements, too. Here is simply a roundup of the announcements that got our attention. TechCrunch volition proceed to update this nonfiction done the extremity of AWS re:Invent truthful beryllium definite to cheque back.

An AI grooming spot and Nvidia compatibility

AWS introduced a new mentation of its AI grooming chip called Trainium3 on with an AI strategy called UltraServer that runs it. The tl;dr: this upgraded spot comes with immoderate awesome specs, including a committedness of up to 4x show gains for some AI grooming and inference portion lowering vigor usage by 40%.

AWS besides provided a teaser. The unreality supplier already has Trainium4 successful development, which volition beryllium capable to enactment with Nvidia’s chips.

Expanded AgentCore capabilities

AWS announced new features successful its AgentCore AI cause gathering platform. One diagnostic of enactment is Policy successful AgentCore, which gives developers the quality to much easy acceptable boundaries for AI agents.

AWS besides announced that agents volition present beryllium capable to log and retrieve things astir their users. Plus it announced that it volition assistance its customers measure agents done 13 pre-built valuation systems.

A non-stop AI cause idiosyncratic bee

AWS announced 3 caller AI agents (there is that word again) called “Frontier agents” including one, called “Kiro autonomous agent” that writes codification and is designed to larn however a squad likes to enactment so it tin run mostly connected its ain for hours oregon days.

Another of these caller agents handles information processes similar codification reviews, and the 3rd does DevOps tasks such as preventing incidents when pushing caller codification live. Preview versions of the agents are disposable now.

New Nova models and services

AWS is rolling retired four caller AI models wrong its Nova AI exemplary household — 3 of which are substance generating and 1 that tin make substance and images.

The institution besides announced a caller work called Nova Forge that allows AWS unreality customers to entree pre-trained, mid-trained oregon post-trained models that they tin past apical disconnected by grooming connected their ain proprietary data. AWS’s large transportation is flexibility and customization.

Lyft’s statement for AI agents

The ride-hailing institution was among galore AWS customers that piped up during the event to stock their occurrence stories and grounds of however products affected their business. Lyft is utilizing Anthropic’s Claude exemplary via Amazon Bedrock to make an AI cause that handles operator and rider questions and issues.

The institution said this AI cause has reduced mean solution clip by 87%. Lyft besides said it has seen a 70% summation successful operator usage of the AI cause this year.

An AI Factory for the backstage information center

Amazon besides announced “AI Factories” that allows large corporations and governments to tally AWS AI systems successful their ain information centers.

The strategy was designed successful concern with Nvidia, and includes some Nvidia’s tech and AWS’s. While companies who usage it tin banal it with Nvidia GPUs, they tin besides opt for Amazon’s newest homegrown AI chip, the Trainium3. The strategy is Amazon’s mode of addressing information sovereignty, oregon the request of governments and galore companies to power their information and not stock it, adjacent to usage AI.

Check retired the latest reveals connected everything from agentic AI and unreality infrastructure to information and overmuch much from the flagship Amazon Web Services lawsuit successful Las Vegas. This video is brought to you successful concern with AWS.