AWS re:Invent 2025 Notes

AWS re:Invent 2025 Notes

AWS re:Invent 2025 held in Las Vegas from December 1st to 5th, is AWS’s premier learning conference for the global cloud computing community. This year’s event marks a decisive shift toward agentic AI—with nearly every major announcement centered on building, deploying, and operating AI agents at scale.

Daily Coverage

DayDateThemeHighlights
Day 1Dec 2CEO KeynoteTrainium3, AgentCore, Nova 2, S3 Vectors, Lambda Durable Functions
Day 2Dec 3Agentic AIStrands SDK, Bedrock RFT, AWS Active Defense, CloudFront Flat Rate
Day 3Dec 4Infrastructure & ClosingGraviton5, Lambda Managed Instances, Werner’s Renaissance Developer

Key Takeaways

Agentic AI Era

  • Agentic AI is the focus - Every major announcement centers on building, deploying, and operating AI agents
  • Framework agnostic - AgentCore, Strands SDK, and Kiro work with any model/framework
  • Enterprise modernization - AWS Transform provides specialized agents for Windows, mainframe, VMware migrations

Custom Silicon Strategy

  • Trainium3 - First 3nm AWS AI chip with 4.4x performance improvement
  • Graviton5 - 192-core ARM processor, 25-35% faster than previous gen
  • Trainium4 - Already announced, continuing aggressive silicon roadmap
  • Project Mantle - Bedrock architecture redesigned specifically for LLM inference

Infrastructure Innovation

Developer Evolution

Werner Vogels’ final re:Invent keynote introduced the Renaissance Developer framework—5 principles for developers thriving alongside AI:

  1. Be Curious - Curiosity leads to learning and invention
  2. Systems Thinking - Think in systems, not isolated parts
  3. Communication - Clearer specs reduce mistakes
  4. Ownership - You build it, you own it
  5. Polymath - Broaden your “T”

“The work is yours, not the tool’s.” — Werner Vogels

Dynamic AI Tooling

  • Kiro Powers - On-demand loading of specialized AI capabilities through MCP
  • Spec-driven development - Natural language requirements → AI implementation
  • Context-aware integration - Tools load only when relevant, preventing context overflow

Major Announcements Summary

AI/ML & Foundation Models

AnnouncementCategoryDescription
Trainium3 UltraServersSilicon3nm, 2.52 PFLOPs, 144 GB HBM3e
Amazon Bedrock AgentCorePlatformDeploy & operate agents at scale
Strands SDKOpen SourceMulti-agent AI systems SDK
Nova 2 FamilyModelsLite, Pro, Sonic, Canvas, Reel
Nova ForgeServiceBuild custom frontier models
Nova ActAgentBrowser-based UI automation
Bedrock RFTTraining66% accuracy gains via reinforcement fine-tuning

Compute & Infrastructure

AnnouncementCategoryDescription
Graviton5 (M9g)Compute192 cores, 25-35% faster
Lambda Managed InstancesServerlessLambda on EC2 in your account
Lambda Durable FunctionsServerlessYear-long workflows
ECS Express ModeContainersRapid containerized app deployment
EKS CapabilitiesKubernetesManaged ArgoCD, ACK, kro

Storage & Data

AnnouncementCategoryDescription
S3 Vectors GAStorageNative vector storage, 90% cost reduction
S3 Tables TieringStorageIntelligent tiering for tables
S3 50TB ObjectsStorageMax object size increased
OpenSearch GPUSearchGPU-accelerated vector search

Networking & Security

AnnouncementCategoryDescription
AWS InterconnectNetworkMulticloud private connectivity
PrivateLink Cross-RegionNetworkCross-region VPC endpoints
CloudFront Flat RateCDNPredictable pricing, no overages
AWS Security AgentSecurityProactive app security
AWS DevOps AgentOperationsAutonomous incident response

Developer Tools

AnnouncementCategoryDescription
KiroAgentAutonomous development agent
Kiro PowersPlatformDynamic AI context loading
SageMaker ServerlessMLServerless model customization
HyperPod CheckpointlessMLMinutes vs hours recovery

Resources

Product Pages

From Werner’s Keynote


Read the Full Coverage

Comments

Kevin Duane

Kevin Duane

Cloud architect and developer sharing practical solutions.