AWS re:Invent 2025 delivered one of the most significant waves of cloud innovation in the event’s history, unveiling more than 40 announcements that redefined what’s possible across AI, data, compute, security, and enterprise operations. From powerful new capabilities in Amazon Bedrock and fully autonomous agent platforms, to breakthrough storage intelligence, next-generation support plans, cost-saving models, and advanced cloud-native services, AWS showcased a future where AI is deeply embedded in every layer of the modern tech stack. This year’s announcements mark a decisive shift toward proactive, AI-driven operations and open the door for organizations of all sizes to build faster, run smarter, and innovate without limits.
Amazon Nova 2 Lite — fast, cost-effective reasoning model for general AI workloads
AWS introduced Nova 2 Lite, a lighter-weight reasoning model available through Amazon Bedrock. Nova 2 Lite brings strong performance for everyday tasks — from code generation and business-process automation to document and data analysis — while optimizing for cost and speed. For tasks needing deeper reasoning, users can optionally enable an extended “thinking” mode and choose among low, medium, or high thinking budgets to balance speed, intelligence, and cost.
Amazon Nova 2 Sonic — new speech-to-speech model for natural voice applications
Among the new Nova models, Nova 2 Sonic adds native speech-to-speech capabilities to the Nova family, enabling real-time voice conversations, natural language interaction, and generative-AI voice applications. This opens doors for voice assistants, voice-driven workflows, multilingual support, and richer conversational UX on Bedrock-powered apps.
Amazon Bedrock expands with 18 fully managed open-weight models including Mistral Large 3 and Ministral 3
AWS announced general availability of 18 new fully managed open-weight models on Bedrock — including Mistral Large 3 and the Ministral 3 series (3B, 8B, 14B). This gives developers and organizations access to state-of-the-art open models that can be used out-of-the-box for a variety of tasks like NLP, reasoning, multimodal workloads, and custom fine-tuning — with the convenience of a fully managed service.
Amazon Bedrock AgentCore gets robust policy controls and quality evaluation tools for trusted AI agents
To help enterprises safely deploy AI agents at scale, AWS enhanced Bedrock AgentCore with two major new capabilities: policy controls and agent performance evaluations. Policy controls allow administrators to define exactly what actions agents can take (tool access, permissions, constraints) — enforcing boundaries before any tool call runs. Meanwhile, built-in evaluation tooling continuously monitors agents for correctness, helpfulness, safety, and compliance, enabling real-time quality insights and governance for agent behavior.
Amazon S3 Vectors now generally available — optimized storage and querying for vector data
AWS made S3 Vectors generally available, delivering a scalable, production-grade vector storage solution directly within S3. This brings native support for vector data (embeddings, AI features, similarity search) to the heart of AWS storage, enabling companies to store and query vectors at a lower cost compared to dedicated vector-database services — a compelling option for AI search, recommendation systems, and embedding workloads.
New AI-driven support plans — shifting AWS Support from reactive help to proactive operations
AWS revealed updated and expanded support plans that integrate AI-powered features with expert guidance, enabling proactive issue detection and prevention rather than reactive troubleshooting. These plans offer faster response times, contextual support, tailored workload optimization, and a higher level of operational reliability for businesses of all sizes. This marks a strategic shift in cloud support aimed at better aligning with modern, complex, and dynamic cloud workloads.
Expanded re:Invent 2025 Highlights
Analytics & Data
AWS Clean Rooms — Launch of privacy-enhancing dataset generation for ML. Clean Rooms lets teams create synthetic datasets for collaborative ML training while preserving privacy, using configurable noise levels and methods to prevent re-identification.
Amazon S3 Vectors (GA) — Amazon S3 now supports production-grade vector storage for embeddings, AI-powered search, and similarity workloads, bringing vector data handling to the core of AWS storage infrastructure.
Compute & Serverless
AWS Lambda Managed Instances — A new model letting Lambda workloads run on EC2 infrastructure while preserving serverless ease. This enables use of specialized compute hardware or instance types with EC2-style pricing, reducing trade-offs between flexibility and cost.
Containers & Orchestration
Amazon EKS — new capabilities for workload orchestration. AWS upgraded its Kubernetes service to simplify orchestration and resource management, easing deployment of containerized workloads with enterprise-grade reliability, security, and automation.
Networking & DNS
Amazon Route 53 Global Resolver (Preview) — A unified anycast-based DNS resolver for hybrid and multi-region architectures. Global Resolver supports both public and private domains globally, helping enterprises simplify DNS management for distributed infrastructures.
Partner & Marketplace Tools
AWS Partner Central — Now accessible directly through the AWS Management Console. This centralizes partner management, solution listings, marketplace management, and partner collaboration through a unified, secure interface.
Identity & Access
IAM Policy Autopilot — An open-source MCP server that helps developers automatically generate valid IAM policies by analyzing code. It reduces manual policy errors and accelerates secure deployments by creating correctly scoped permission sets.
Support & Operations
New AI-powered AWS Support plans — AWS announced revamped support offerings combining AI-driven tooling with human expert guidance. New plans introduce faster response times, proactive issue detection, and workload optimization recommendations — shifting support from reactive troubleshooting to proactive operations.
Application Modernization
AWS Transform — full-stack Windows modernization: Extended to support entire application stacks, including database (e.g., SQL Server → Aurora PostgreSQL), UI migration (like ASP.NET Web Forms to Blazor), and cross-platform deployment (e.g., Linux/containers). This aims to speed up modernization up to 5× for classic Windows apps.
AI & Generative AI Services
Expanded open-weight models on Amazon Bedrock — Addition of 18 new fully managed open-weight models to Bedrock, offering developers broad choice for reasoning, NLP, multimodal tasks, and fine-tuning, without managing model infrastructure themselves.
Enhanced Bedrock data processing & retrieval — Upgrades to Bedrock to support better structured querying, multimodal processing, and graph-style modeling, making it easier to build advanced AI applications.
Multimodal toxicity detection in Bedrock Guardrails (preview) — Bedrock’s safety and moderation stack now supports image-based as well as text-based toxicity detection for safer generative AI deployments.
New ML-friendly tooling: Amazon SageMaker HyperPod recipes — Prebuilt recipes to quickly train or fine-tune large foundation models (e.g., LLama 3.1 405B) in minutes — lowering the barrier for custom model training and experimentation.
Business-user ML with Amazon Q in QuickSight & Q Developer — New “Amazon Q in QuickSight” adds fast, spreadsheet-like scenario analysis for business users. Q Developer enables non-ML experts to build ML models with natural-language instructions — accelerating democratization of ML and analytics.
AWS Announcements — December 2, 2025
AI, Agents & Bedrock
| Announcement | Summary |
|---|---|
| Amazon Bedrock AgentCore – New Capabilities | Adds policies, episodic memory, agent evaluations, and bidirectional streaming to make agents safer, more autonomous, and production-ready at scale. |
| 18 New Open-Weight Models on Amazon Bedrock | Introduces new open-source reasoning, multimodal, and coding models, dramatically expanding developer choice. |
| Bedrock Guardrails: Multimodal Toxicity Detection (Preview) | Enables detection of harmful or inappropriate content in text, images, and multimodal inputs for safer AI applications. |
| Amazon Bedrock Data Intelligence Updates | Enhancements include richer graph retrieval, entity extraction, document structuring, and improved RAG grounding. |
| Amazon SageMaker HyperPod Recipes | One-click training templates for massive LLMs (like Llama 3.1–405B) simplify distributed training setups. |
| Amazon Q in QuickSight | Adds conversational BI modeling, “spreadsheet-style” scenario planning, and advanced analytical automation. |
| Amazon Q Developer: ML Model Builder | Low/no-code ML creation using natural language, enabling business users to generate models without data science expertise. |
| Amazon Q Apps – New Enhancements | Adds more powerful workflow chaining, context retention, and advanced automations for building custom AI apps. |
| Amazon Q Business – Expanded Skills | Improves organizational Q&A, metrics analysis, summarization, and document insight capabilities. |
| Amazon Kendra + Q Improvements | Better semantic search and hybrid reasoning across structured + unstructured data sources. |
Join the conversation! Your thoughts help the community grow.