AWS re:Invent 2025: A technical deep dive into the new AI-native cloud era and what it means for JAM clients

AWS re:Invent 2025 marked a decisive shift from traditional cloud operations to AI-native infrastructure, agent-driven automation and frontier-scale model development.

For engineering leaders, SRE teams, and platform owners, the announcements represent a fundamental redesign of how applications will be built, deployed, secured, and operated over the next decade.

At Just After Midnight, we support mission-critical systems for digital, SaaS and enterprise organisations around the world. The depth and direction of AWS’s new capabilities align directly with the challenges our clients face: scaling AI workloads, improving reliability under increased system complexity, reducing operational toil and moving toward automated infrastructure and application operations.

Below is a technical breakdown of the most consequential AWS launches from re:Invent and how they will impact the architectures and operations we build and run for clients.

1. New foundation models and AI capabilities

AWS expanded the Amazon Nova family with four significant additions, each designed for different workloads and reasoning patterns.

Amazon Nova 2 Lite

A lightweight reasoning model with fast performance, cost efficiency, and a million-token context window. Optimised for everyday enterprise tasks.

Amazon Nova 2 Sonic

A high quality speech-to-speech model capable of multilingual conversations, dynamic speech control, and telephony-grade interaction. Ideal for customer support and real time communications.

Amazon Nova 2 Omni (Preview)

A multimodal model supporting text, images, video, and speech inputs with both text and image outputs. This correct name replaces the earlier “AMI” reference.

Amazon Nova Act

A new service for building agent-driven UI automation workflows with over 90 percent reliability. Supports browser based task automation across form filling, navigation, search extraction, shopping and booking flows, and QA testing.

Amazon Nova Forge

A major milestone. Nova Forge enables organisations to build custom frontier models by mixing Nova training data with their own proprietary datasets. The output is a domain specialised model without the cost and complexity of training from scratch.

Amazon Bedrock model updates

AWS added 18 fully managed open-weight models including

• Mistral Large 3
• Ministral 3 (3B, 8B, 14B)
• Models from Google, Kimi AI, MiniMax AI, NVIDIA, OpenAI, and Qwen

Bedrock also introduced reinforcement fine tuning to improve model accuracy by 66 percent without large labeled datasets.

Impact for JAM clients:

These model capabilities create new opportunities to enhance SRE intelligence, automated diagnostics, customer support AI, log understanding, documentation analysis, and predictive incident detection.

2. Agent platforms and autonomous operations

AWS is clearly positioning the cloud for enterprise-scale agentic systems.

Amazon Bedrock AgentCore

AgentCore now includes

• Quality evaluations
• Policy controls
• Improved memory and context
• Enhanced conversation capabilities

This standardises policies and evaluation frameworks for agent workflows, closing a major gap in safety and predictability.

AWS DevOps Agent (Preview)

A new autonomous agent designed to act as an on-call SRE. It can

• Analyse telemetry across CloudWatch, GitHub, ServiceNow, and other systems
• Identify root causes
• Propose fixes
• Coordinate resolution

AWS Security Agent (Preview)

A proactive security agent capable of

• Design reviews
• Code analysis
• Contextual penetration testing
• Secure development workflow integration

Impact for JAM clients:

JAM’s SRE and managed services teams will be able to integrate AgentCore, DevOps Agent, and Security Agent into incident management pipelines, observability stacks, and automated change validation flows, enabling faster recovery and more autonomous operations.

3. AI infrastructure and training enhancements

Amazon EC2 Trn3 UltraServers

Powered by AWS Trainium3 chips.

• Up to 4.4 times higher performance
• Up to 4 times better performance per watt
• Designed for frontier-scale model training
• Native PyTorch integration

Amazon EC2 X8aedz Instances

Memory optimised instances powered by 5th Gen AMD EPYC processors.

• Up to 5 GHz clock
• 3 TiB of memory
• Ideal for EDA workloads, large databases, and high frequency compute.

Amazon SageMaker AI Capabilities

• Serverless MLflow for rapid experimentation
• Checkpointless and elastic training on SageMaker HyperPod
• Serverless customization for faster fine tuning

Impact for JAM clients:

These capabilities reduce the cost and complexity of training and inference workloads, making large scale AI operations viable for SaaS applications, media workloads, bioinformatics, and financial modelling.

4. Modernisation and migration at scale

AWS expanded the AWS Transform service family.

AWS Transform Custom

AI powered code modernisation that learns organisational patterns and automates transformations across repositories.

AWS Transform full-stack Windows modernization

Coordinated transformation of Windows applications.

• Across code
• UI frameworks
• Databases
• Deployment configurations

AWS Transform for mainframe

Reimagine capabilities and automated test generation for converting legacy mainframe applications to cloud native architectures.

Impact for JAM clients:

Modernisation timelines drop from years to months, transforming legacy systems into scalable, supportable, cloud native architectures that JAM can then manage 24 hours a day.

5. Security, identity and compliance

Amazon GuardDuty Extended Threat Detection

Now supports both Amazon EC2 and Amazon ECS.

Provides unified visibility across VM and container attack paths.

AWS Security Hub GA

Includes:

• Near real-time analytics
• Risk prioritisation

IAM Policy Autopilot

New open source MCP server for AI assistants to generate accurate IAM policies.

Impact for JAM clients:

Security posture becomes more automated, traceable, and standardised, improving reliability of 24-hour operations.

6. Storage, data and observability improvements

Amazon S3 Vectors GA

• Up to 2 billion vectors per index
• 100 ms query latencies
• 90 percent lower cost than specialised vector databases

Amazon FSx for NetApp ONTAP S3 integration

Enables direct S3 access to file system data.

Amazon S3 Tables

• Now supports replication
• Now supports Intelligent Tiering

Amazon S3 Storage Lens

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

Amazon CloudWatch unified data management and analytics

Normalises operational, security, and compliance data across sources with built-in analytics.

Amazon OpenSearch Service GPU acceleration and auto-optimization

Ten times faster vector indexing at one quarter the cost.

Impact for JAM clients:

JAM can design next-generation observability and data pipelines where vector embeddings, unified telemetry, and ML-powered analysis become first-class components of the operational stack.

7. Application, networking and serverless enhancements

AWS Lambda Managed Instances

Run Lambda functions on EC2 capacity while maintaining serverless operational models.

AWS Lambda Durable Functions

Multi-step workflows that can run up to one year.

Amazon EKS Capabilities

Managed workload orchestration and cloud resource management.

Amazon Route 53 Global Resolver (Preview)

Unified hybrid DNS management with anycast resolution.

Impact for JAM clients:

These features simplify distributed architectures, multi-region patterns, and long-running workflow implementations across SaaS and enterprise platforms.

8. AWS AI Factories

A major global infrastructure shift.

AWS AI Factories let organisations deploy fully managed AWS AI infrastructure inside their own data centers.

Includes:

• Foundation models
• Trainium
• Specialised hardware
• AWS services

Impact for JAM clients:

Opens a new path for AI adoption in industries with strict sovereignty demands while maintaining fully managed operability.

Conclusion? It’s the start of the AI-native cloud era

AWS has moved beyond incremental cloud enhancements and is now building a first principles architecture for AI-native engineering, operations, observability, and application development.

For JAM clients, this translates into:

• More autonomous incidents and operations
• Faster model training and smarter inference
• Simpler modernisation of legacy applications
• Deeper observability and telemetry intelligence
• Stronger security validation at scale
• New hybrid and sovereign AI deployment options
• Reduced operational toil and increased reliability

JAM will integrate these technologies into our SRE, observability, and managed cloud services throughout 2025 and 2026, giving our customers a path to operate confidently in the next era of cloud computing.

If you’re planning your next phase of cloud or AI modernisation, our team can walk you through what’s possible and where to start. Just get in touch.