At AWS re:Invent 2025 in Las Vegas, CEO Matt Garman declared a sweeping vision for Amazon Web Services — one that stakes the company’s future on artificial intelligence infrastructure and a world full of AI agents. From new chips to hybrid data-center models and aggressive bets on agentic workflows, Garman made clear how AWS intends to anchor enterprise AI over the coming years.
Garman began by framing AWS as a juggernaut: a $132 billion business growing at 20% year over year, with an absolute increase of “about $22 billion” in the past 12 months — a sum larger than the annual revenue of more than half of the Fortune 500, he said. Garman told the crowd of over 60,000 live attendees, along with nearly 2 million online viewers, that the growth is “coming from across the business,” from basic storage to advanced AI services.
Click the banner below for deeper insight into IT modernization.
From Infrastructure to Agents: AWS’s Vision for AI at Work
According to Garman, AWS is eyeing a paradigm shift, targeting not incremental gains with AI. He said that while many organizations are still experimenting with AI assistants or chatbots, “the true value of AI has not yet been unlocked.” The breakthrough will come with AI agents — autonomous systems that can act on behalf of people and organizations.
“We believe the advent of AI agents has brought us to an inflection point,” he said. “In the future, there’s going to be billions of agents inside of every company, across every imaginable field.” From healthcare to payroll processing to customer service, agents promise to multiply human productivity exponentially.
Turning that vision into reality will require massive new infrastructure, he said: “You need a highly scalable and secure cloud that delivers the absolute best performance for your AI workloads at the lowest possible cost. Only AWS is building across every layer of silicon, software, networking and data centers to meet that demand.”
AWS Takes Big Steps in Advancing AI Infrastructure
A central theme of Garman’s remarks was silicon: AWS’ own chips, along with the latest GPUs, underpinning massive AI workloads. On that front, he made several announcements:
- Trainium3, AWS’ first 3‑nanometer AI chip, is now available for customers. Each fully equipped AWS UltraServer contains 144 Trainium3 chips, which AWS says delivers more than four times the computing power and better energy efficiency compared with the previous generation. Trainium3 is built not only for training AI models but also for running them at scale, Garman said. Early users, including major AI companies, report they’ve been able to cut training and inference costs by up to half, he said.
- AWS also introduced new high-performance GPU-powered servers designed to handle very large AI workloads efficiently as part of its partnership with NVIDIA.
- For companies that want AI infrastructure on‑premises, AWS launched AWS AI Factories. This allows organizations to run dedicated, AWS-managed AI hardware in their own data centers. Customers can also access AWS services such as Bedrock and SageMaker while keeping control over security, compliance and data ownership.
Taken together, these updates reflect AWS’ belief that future enterprise AI workloads will demand not just GPU clouds but flexible hybrid deployment models combining custom silicon, cloud reliability and on‑premises control.
Click the banner below to learn how enterprises are unlocking artificial intelligence’s potential.
Generative AI at Enterprise Scale: Bedrock, Nova and Model Choice
On the software side, AWS continues to push its generative AI stack forward. Garman said that for many customers, the first step isn’t building a model from scratch but deciding which foundation model to use, and when to customize. Rather than banking on a single model, AWS is making a wide selection available: large and small, general-purpose and specialized. Over the past year, the number of models available in Amazon Bedrock has nearly doubled.
AWS’ own model suite, Amazon Nova, is expanding. Garman said that recent updates are designed for enterprise workloads that demand cost‑effective, low-latency reasoning, speech, image and video capabilities.
Garman repeatedly invoked a slogan: “Giving you all the freedom to invent.” For AWS, that means giving developers the tools to build AI applications at a previously unthinkable scale.
What stands out about re:Invent 2025 is how AWS positions itself not merely as a cloud-infrastructure provider but as the backbone of the next generation of AI, delivering chips, servers, networking, models and deployment options. That ambition reflects a belief that AI’s future will be agentic, distributed and infrastructure-heavy.
“We want to reimagine every single process in the way that all of us work,” Garman said.
