AWS bets big on agentic AI and supercharged compute

AI factories and data sovereignty

To meet government and industry requirements for local data control, AWS launched AI Factories—private, dedicated AWS regions that operate within customers’ data centres. These facilities grant access to leading AI infrastructure such as Amazon SageMaker and Amazon Bedrock while ensuring that sensitive data remains under domestic oversight. The concept targets sectors such as banking, healthcare, and the public sector.

Expanding the Bedrock ecosystem

Amazon Bedrock, AWS’s fully managed AI platform, now supports more than 100,000 organisations worldwide. Its new Nova 2 family of models—Lite, Pro and Omni—covers tasks from cost-efficient text generation to advanced reasoning and multimodal analysis across text, image, audio, and video. Garman emphasised that AWS is committed to providing a wide selection of models so customers can prioritise cost, speed, or capability according to their needs.

From generative to agentic AI 

Dr. Sivasubramanian described agentic AI as the logical successor to today’s generative AI. Through Amazon Bedrock AgentCore, developers can design and supervise AI agents that operate independently inside secure cloud environments. AgentCore includes mechanisms for short-term and long-term memory so agents can learn from previous experiences and improve over time. Organisations such as Sony Group and Blue Origin are already employing this technology to automate compliance workflows and streamline engineering simulations.

Customising models for every enterprise

AWS launched Amazon Bedrock Reinforcement Fine-Tuning (RFT) to simplify model optimisation. The new system enables customers to improve accuracy through guided feedback rather than manual retraining. For deeper personalisation, Amazon Nova Forge lets businesses combine proprietary data with AWS datasets to produce domain-specific “Novella” models. These private models capture institutional knowledge while protecting sensitive information.

Building trust through verification

Security and reliability featured heavily throughout the event. AWS showcased its work in automated reasoning, which uses mathematical logic to verify that systems behave as expected. The approach will extend to AI agents, ensuring they act predictably and remain within enterprise policies. Sivasubramanian said these capabilities will make it easier for companies to adopt autonomous systems responsibly.

Frontier agents to accelerate development

WS expanded its internal use of AI with a new set of Frontier Agents: the Kiro Autonomous Agent for software development, AWS Security Agent for vulnerability detection, and AWS DevOps Agent for cloud operations. Early pilots inside Amazon showed significant productivity gains, including one engineering project completed in less than three months instead of 18. The company sees this as evidence that human-machine collaboration can reshape software delivery.

Asia’s momentum in AI adoption

AWS expects strong regional demand for its new agentic offerings. AI Factories help Asia-Pacific governments and corporations meet local compliance while maintaining access to global-scale resources. Nova Forge supports the development of local-language models and culturally specific applications. Garman highlighted Asia’s rapid digital transformation and said the region’s appetite for secure cloud innovation positions it as one of AWS’s fastest-growing AI markets.

A platform evolving with intelligence

re:Invent 2025 marked AWS’s transition from a traditional infrastructure supplier to a fully integrated AI platform.

With Amazon Bedrock orchestrating models, AgentCore managing agents, and Amazon SageMaker training new systems, the company is building a foundation for self-learning software. The Las Vegas event made clear that the cloud is no longer just a place to run applications—it is becoming an active participant in creating them. For enterprises, the challenge ahead is how quickly they can let their agents start building on their behalf.