At the AWS re:Invent 2025 event in Las Vegas, Amazon Web Services unveiled its ambitious vision focused on agentic AI, an evolution that could transform how businesses leverage artificial intelligence. With over 60,000 attendees on-site and millions connected online, AWS showcased innovations aimed at propelling the use of autonomous agents in the professional world. Discover how AWS intends to redefine the technological landscape with its new tools and infrastructures.
The 3 must-know facts
- AWS highlights agentic AI as the next major step in artificial intelligence.
- The new Trainium3 chip powers AWS servers for unprecedented agentic workloads.
- Nova models are integrated into Bedrock, optimized for Amazon hardware.
A turning point with agentic AI
At re:Invent 2025, AWS highlighted its vision for the future of AI by focusing on agentic AI. CEO Matt Garman shared that AI agents, capable of executing and automating tasks, represent the next big step in digital transformation. This approach aims to provide “material value” to businesses by automating complex tasks rather than merely reacting to commands.
Dr. Swami Sivasubramanian, VP of Agentic AI at AWS, illustrated this evolution by comparing traditional chatbots to fully autonomous agents. While chatbots inform about potential issues, agents take the necessary actions to resolve them. This shift offers developers the freedom to build proactive systems.
Hardware and infrastructure innovations
AWS unveiled Trainium3, a chip designed to power Ultraservers, enabling the management of agentic workloads on an unprecedented scale. This technological advancement is essential to meet the growing computational power needs of AI. The Ultraservers, combining 144 chips into a single core, promise impressive computing performance with hundreds of petaflops of power.
In parallel, AWS introduced the AI Factory, a solution allowing companies to manage their own data centers using Trainium clusters, Nvidia GPUs, and services like Bedrock and SageMaker. This initiative addresses data sovereignty concerns, particularly in Europe.
Nova models and diversity of AI solutions
The AWS Nova model family has been expanded to offer solutions optimized for Amazon hardware. Nova 2 offers variants for reasoning, conversation, and efficient inference, integrated into the Bedrock platform. This diversity allows companies to combine different models to meet various needs, a trend also noted by Jeff Boudier of Hugging Face.
AWS also launched Nova Forge, a platform facilitating access to Nova model checkpoints. This allows companies to modernize and adapt their workloads using their own data to create customized agents.
AWS context and history
Amazon Web Services, a subsidiary of Amazon, is one of the leading cloud service providers globally. Launched in 2006, AWS quickly evolved to offer a wide range of cloud services, including storage, computing, and now artificial intelligence. With clients ranging from startups to large enterprises, AWS has established itself as a key player in the digital transformation of businesses. AWS’s strategy is based on continuous innovation and adaptation to the changing needs of the market, as evidenced by its current commitment to agentic AI technologies.