Skip to main content

AWS re:Invent 2025 - Major Announcements

AWS re:Invent 2025 marked a paradigm shift from incremental cloud improvements to fundamental architectural transformations. These announcements redefine autonomous operations, serverless boundaries, cost efficiency, and foundational infrastructure capabilities.

1. AWS Transform (Autonomous Tech Debt Remediation)​

For decades, cloud modernization required hiring massive systems-integrator teams spending years rewriting legacy systems. AWS Transform breaks this paradigm with agentic AI-driven autonomous modernization.

Architecture Game-Changer:

  • Acts as 24/7 autonomous engineer
  • Full-stack modernization of entire legacy portfolios
  • Auto-converts COBOL mainframes to cloud-native Java
  • Refactors .NET applications
  • Migrates massive VMware workloads
  • Up to 5x faster than traditional approaches

Continuous Remediation:

  • Autonomously detects architectural tech debt
  • Prioritizes fixes at scale
  • Patches as developers push code
  • Cuts execution time by up to 80%

Capabilities:

  • Windows application modernization (5x faster)
  • Mainframe reimagination with automated testing
  • Reduces modernization timelines from years to months

2. AWS Lambda Managed Instances​

Historically, architects faced a binary choice: Lambda's simplicity with hardware limitations, or EC2/ECS with massive management overhead.

Architecture Game-Changer:

  • Run serverless Lambda workloads on specialized EC2 compute clusters
  • Retain serverless simplicity (AWS manages scaling, patching, availability)
  • Gain access to specialized GPU/AI chips
  • Use unique EC2 pricing models
  • Leverage specific hardware optimizations

3. AWS Lambda Durable Functions​

Native stateful orchestration and automatic checkpointing directly in Lambda eliminates complex Step Function management.

Capabilities:

  • Build applications coordinating multiple steps reliably
  • Duration: seconds to up to one year
  • Pause, resume, track state natively within serverless containers
  • Multi-step processes without external databases
  • Long-running AI agent workflows

Impact:

  • Complex business logic with built-in retries
  • State management directly in Lambda
  • Reduced reliance on external orchestrators

4. Native Cross-Cloud High-Speed Interconnects (AWS-GCP)​

AWS and Google Cloud launched native, open-standard, private high-speed interconnect breaking multi-cloud barriers.

Architecture Game-Changer:

  • Eliminated complex third-party network providers (Megaport)
  • Removed massive ingress/egress financial penalties
  • Native multi-cloud architecture support

Benefits:

  • Seamlessly split workloads between clouds
  • Store datasets in AWS S3 data lake
  • Run real-time analytics on Google Cloud infrastructure
  • Ultra-low latency
  • Minimized data-transfer friction

5. Custom Next-Gen AI Silicon​

AWS Graviton5​

Fifth-generation CPU providing best price-performance for broad workload range in EC2.

Performance:

  • 25% better raw compute performance vs previous generation
  • Radical energy efficiency enhancements
  • 192 cores (M9g instances)

AWS Trainium3​

Deployed via Trn3 UltraServers using 3nm architecture.

AI Compute Capabilities:

  • 4x performance boost for multi-billion parameter AI models
  • 30-40% better price-to-performance ratio vs standard GPUs
  • Makes massive AI workloads financially viable for corporate budgets
  • First 3nm AI chip from AWS

Complementary Hardware:

  • NVIDIA GB300 UltraServers on EC2
  • Massive on-demand scale for high-throughput inference

6. Amazon S3 Enhancements​

50TB Object Size Limit​

Increased from 5TB to 50TB per individual object.

Impact:

  • Fundamentally changes AI training pipeline data ingestion
  • Massive raw media dataset handling
  • Eliminates complexity of splitting/tracking/reconstructing multi-part files

S3 Vectors (Generally Available)​

Vector storage and querying built directly into S3.

Capabilities:

  • Support up to 2 billion vectors per index
  • 100ms query latencies
  • Reduces RAG costs by up to 90% vs specialized databases
  • Pulls AI data gravity to fundamental storage layer

S3 Tables:

  • Replication support
  • Intelligent-Tiering

S3 Storage Lens:

  • Performance metrics
  • Support for billions of prefixes
  • Export to S3 Tables

7. Autonomous Frontier Agents​

AI shifting from request-response APIs to persistent autonomous actors.

AWS Frontier Agents​

Kiro Autonomous Agent:

  • Virtual developer working autonomously for days
  • Manages state
  • Reviews architectural designs
  • Fixes bugs without human intervention

AWS Security Agent (Preview):

  • AI-powered design reviews
  • Code analysis
  • Contextual penetration testing

AWS DevOps Agent (Preview):

  • Acts as autonomous on-call engineer
  • Analyzes CloudWatch, GitHub, ServiceNow data

Agentic Amazon Connect​

Specialized agent-driven workflows:

  • Connect Decisions: Supply chain automation
  • Connect Talent: Hiring workflows
  • Connect Health: Healthcare operations
  • Turns AI into proactive operational teammates

Amazon Quick​

Desktop AI assistant:

  • Generates visual assets
  • Builds custom dashboards via natural language
  • Integrates with external workspaces

8. Amazon Bedrock & OpenAI Partnership​

Early 2026 partnership expansion reshaping enterprise AI consumption.

GPT-5.5 and Codex on Bedrock​

  • Access state-of-the-art reasoning within AWS security boundary
  • Keep data entirely within AWS trust perimeter
  • Native integration with AWS environments

Bedrock Managed Agents (OpenAI-powered)​

  • Combines OpenAI harness with AWS infrastructure
  • Build production-ready agents without stitching clouds
  • No disparate environment management

Amazon Nova Family​

Nova 2 Sonic (Speech-to-Speech):

  • Multilingual conversations
  • Dynamic speech control
  • Crossmodal inputs
  • Improved telephony integration

Nova 2 Lite:

  • Fast, cost-effective everyday tasks
  • Extended thinking capabilities
  • Built-in tools
  • Million-token context window

Nova Forge:

  • Build custom frontier models
  • Deep domain expertise embedding
  • Without traditional cost/compute/time barriers
  • "Open training" - blend proprietary data with Nova 2 checkpoints

Nova 2 Omni (Preview):

  • All-in-one multimodal reasoning and image generation
  • Text, images, video, speech inputs
  • Text and image outputs

Nova Act (Generally Available):

  • Automate browser-based tasks
  • Form filling, search & extract, shopping & booking, QA testing
  • Over 90% reliability

Bedrock AgentCore​

  • Enhanced policy controls
  • Quality monitoring
  • Improved memory
  • Natural conversation abilities

Model Additions​

  • 18 fully managed open weight models
  • Mistral Large 3 and Ministral 3 (3B, 8B, 14B)
  • Models from Google, Kimi AI, MiniMax AI, Mistral AI, NVIDIA, OpenAI, Qwen

Reinforcement Fine-Tuning​

  • 66% accuracy gains over base models
  • No large labeled datasets required
  • No deep ML expertise needed

9. Compute & Containers​

Amazon EC2 Enhancements​

X8aedz Instances:

  • 5th Gen AMD EPYC processors
  • Up to 5 GHz processor speeds
  • 3 TiB memory

Amazon EKS​

  • Fully managed platform capabilities
  • Workload orchestration
  • Cloud resource management

10. Database Innovations​

Database Savings Plans​

New pricing model across AWS Databases.

Amazon RDS​

  • Developer Edition support for SQL Server
  • M7i/R7i instance support with optimized CPU
  • Storage expansion up to 256 TiB

11. AI/ML Infrastructure​

Amazon SageMaker AI​

  • Serverless MLflow integration
  • Checkpointless and elastic training on HyperPod
  • Instant recovery from failures
  • Automatic scaling based on resource availability
  • New serverless customization

Amazon OpenSearch Service​

  • GPU acceleration for vector databases
  • 10x faster performance
  • Quarter of the cost
  • Auto-optimization capabilities

12. Security & Governance​

AWS Security Hub (GA)​

  • Near real-time analytics
  • Risk prioritization

Amazon GuardDuty​

  • Extended Threat Detection for EC2 and ECS
  • Unified visibility across VM and container environments

IAM Policy Autopilot​

  • Open source MCP server
  • Analyzes code to generate valid IAM policies

13. Storage & Networking​

Amazon FSx for NetApp ONTAP​

  • S3 integration for seamless data access
  • Access file system data through S3

Amazon Route 53 Global Resolver (Preview)​

  • Unified service resolving public and private domains globally
  • Secure, anycast-based resolution

14. AWS AI Factories​

Deploy fully managed AWS AI infrastructure in your data centers:

  • Pre-integrated foundation models
  • Specialized hardware
  • AWS services on-premises

15. Management & Monitoring​

Amazon CloudWatch​

  • Unified data management and analytics
  • Operations, security, compliance
  • Native OCSF and Apache Iceberg format support

AWS Support Plans​

  • AI-powered insights with expert guidance
  • Faster response times
  • Proactive monitoring (performance, security, cost)

16. AWS Clean Rooms​

Privacy-enhancing synthetic dataset generation:

  • ML model training datasets
  • Preserve statistical patterns
  • Protect individual privacy
  • Configurable noise levels
  • Protection against re-identification