AWS reveals major AI/cloud announcements at re:Invent 2025

• AWS introduces new chips and tools for AI at re:Invent 2025.
• Updates mark a shift from AI pilots to long-term infrastructure planning.

At its re:Invent 2025 conference in Las Vegas, AWS outlined a set of updates aimed at companies building and running AI systems at scale. The announcements focused on custom chips, new tools for AI agents, and changes to how large organisations deploy and manage AI workloads in the cloud.

The updates reflect a growing shift in enterprise AI use. Instead of testing models in isolation, many firms are now trying to run AI in customer service, software development, analytics, and internal operations. AWS’s announcements suggest the company is adjusting its cloud services to support that move from experiments to daily use.

New chips to support larger AI workloads

One of the central themes at re:Invent was infrastructure. AWS introduced Trainium3 Ultra, a new version of its custom AI chip designed for training large models. According to AWS, Trainium3 Ultra systems can be clustered into large groups, allowing companies to train models with hundreds of billions of parameters.

AWS also shared details about Graviton5, the latest generation of its general-purpose processor. While Graviton chips are not built only for AI, AWS said the new version offers higher performance and lower energy use for workloads like data processing, databases, and backend services that often support AI applications.

AWS framed these chips as a way for customers to reduce their dependence on third-party hardware. In its re:Invent news release, the company said the goal is to give customers more control over cost, scale, and availability when running AI workloads.

A push toward AI agents

Another major focus was AI agents. AWS introduced updates to Amazon Bedrock, its service for building applications with foundation models. One of the additions, called AgentCore, is designed to help developers create AI agents that can take actions in systems, not only respond to prompts.

AWS described these agents as tools that can handle tasks like pulling data from internal systems, triggering workflows, or responding to customer requests without constant human input. The company said AgentCore includes controls to define what an agent can access and how it behaves, which is meant to reduce the risk of errors or misuse.

In a keynote address, AWS CEO Matt Garman said, “Customers want AI systems that don’t just generate text, but actually help get work done.”

AI factories and large-scale deployment

AWS also introduced the concept of AI Factories, which it described as pre-built environments that combine compute, storage, networking, and software tools for large AI projects. These are meant for organisations training or running multiple models at once, often in regions.

According to AWS, AI Factories are intended to shorten the time it takes to move from development to production. Rather than assembling infrastructure piece by piece, customers can start with a standard layout and adjust it over time. AWS said this approach is aimed at companies with long-term AI roadmaps, not short-term pilots.

Changes to how enterprises think about AI costs

Cost control was a recurring topic. Training and running large AI models remains expensive, and AWS acknowledged that many customers are struggling to predict or manage those costs. Alongside new chips, AWS pointed to tools that help track use and allocate spending in teams.

AWS said these updates are meant to make AI spending more visible to finance and IT leaders, not just engineers. That reflects a broader change in how companies view AI projects, which are increasingly treated as long-term infrastructure investments not experimental budgets.

Industry analysts have noted that this focus on cost transparency is becoming more important as AI workloads grow. In coverage following re:Invent, several analysts said cloud providers are under pressure to show that AI can be scaled in a controlled way, not only at the high end of spending.

A quieter tone, but a clear direction

Unlike earlier AI events that focused on bold claims or rapid breakthroughs, AWS’s re:Invent 2025 announcements were more measured. The company spent less time on model performance numbers and more time on how AI systems fit into existing enterprise environments.

Many companies are past the stage of asking whether to use AI. The harder questions now are how to run it reliably, how to control access, and how to keep costs in check over time. AWS’s updates suggest it sees its role as providing the underlying systems to support that shift, not pushing a single AI product or model.

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.

TechHQ is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.