AWS re:Invent 2025 — Day 2: Keynotes, Agents, and a Wave of Announcements

Day 2 at AWS re:Invent 2025 — Matt Garman’s keynote, 25 rapid-fire announcements, AI everywhere, and a few personal milestones along the way.

There’s something special about the first keynote of the week. The event feels like it clicks into its next gear — thousands of people settling into seats, the stage lights firing up, and AWS laying down its vision for the year ahead.

Today was Matt Garman’s keynote, and it didn’t take long before it became very clear: 2025 is the year AWS doubles down on AI agents, developer velocity, and the foundations required to make all of that scale.

Image

The Keynote: Matt Garman Finds His Rhythm

Matt walked on stage with noticeably more confidence and presence than last year. Less “guest in someone’s lounge room”, more “I own this stage now”. And he opened with three themes that set the tone:

Security is priority one.
Partners are critical.
Startups are flocking to AWS.

We heard a customer story from last year’s Unicorn Award winner, Audioshake, followed by the keynote’s anchor message:

“The future belongs to developers.”

Heroes and User Groups both received a heartfelt call-out — something I always appreciate as someone deeply invested in the community.

The framing was simple: developers should be able to build anything they imagine without capital, without friction, and without waiting for infrastructure to catch up. Or as Matt repeated several times:

“Why not?”

Why not remove cost barriers?
Why not collapse the time between idea and impact?
Why not make it possible for every organisation to train their own frontier model?

AI Infrastructure: The Foundation Layer

Garman’s first deep dive was the one we all expected — the infrastructure required to run AI at scale.

NVIDIA + AWS continues to tighten

AWS touted its diagnostics, firmware-level tuning, and deep partnership with NVIDIA, announcing:

Amazon EC2 P6e UltraServers (GA)

NVIDIA GB300 NVL72-based clusters designed for high-end training workloads.
https://aws.amazon.com/about-aws/whats-new/2025/12/amazon-ec2-p6e-gb300-ultraservers-nvidia-gb300-nvl72-generally-available/

These follow the earlier P6 family, pairing Blackwell-class performance with AWS’s Nitro fabric.

AWS AI Factories (New)

https://aws.amazon.com/about-aws/whats-new/2025/12/aws-ai-factories/

I’ll be honest — this one feels like a plot twist: AWS delivering AI superclusters inside customer data centres. It’s marketed as a way to accelerate AI buildouts by months or years, bundling Trainium, NVIDIA GPUs, HPC networking, and storage into a drop-in rack.

Useful? Yes.
Surprising from AWS? Also yes.

Trn3 UltraServers for next-gen model training

https://aws.amazon.com/about-aws/whats-new/2025/12/amazon-ec2-trn3-ultraservers/

Trainium 3 delivers faster and cheaper generative AI training, continuing the trend of AWS gradually carving out its own silicon advantage.

Bedrock: Model Choice, Scale and a Very Big Year

We then shifted to Amazon Bedrock, where the usage graphs — “astounding”, as Matt put it — show adoption well beyond early experiments. Bedrock now runs some of the largest AI applications globally.

Fully Managed Open-Weight Models Arrive

https://aws.amazon.com/blogs/aws/amazon-bedrock-adds-fully-managed-open-weight-models/

Newcomers include:

  • Mistral Large 3
  • Ministral 3
  • MiniMax AI
  • Moonshot AI
  • NVIDIA
  • OpenAI
  • Qwen
  • Google models

The theme was clear: model choice remains a core differentiator.

Amazon Nova 2: The Family Expands

Last year’s Nova launch was positioned as a long-term bet. This year, we got the next chapter:

Nova 2 Lite & Nova 2 Pro

https://aws.amazon.com/about-aws/whats-new/2025/12/nova-2-foundation-models-amazon-bedrock/
https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-lite-a-fast-cost-effective-reasoning-model/

  • Lite is positioned as the fast, cost-efficient reasoning model for everyday workloads — great at instruction following, document extraction, and code.
  • Pro is the heavyweight — designed for multi-step, complex agentic workflows.

Nova 2 Sonic

https://aws.amazon.com/about-aws/whats-new/2025/12/amazon-nova-2-sonic-real-time-conversational-ai/
https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-sonic-next-generation-speech-to-speech-model-for-conversational-ai/

A real-time, speech-to-speech model with extended language support aimed squarely at conversational AI.

Nova 2 Omni

A multimodal variant accepting text, image, video, and audio inputs. Ideal for analytical workloads and media understanding.

Nova Forge: Build Your Own Frontier Model

One of the biggest announcements of the keynote.

Amazon Nova Forge (New)

https://aws.amazon.com/about-aws/whats-new/2025/12/amazon-nova-forge-frontier-models-nova/
https://aws.amazon.com/blogs/aws/introducing-amazon-nova-forge-build-your-own-frontier-models-using-nova/

Forge lets you build your own frontier model using the Nova architecture.
The resulting custom models are called Novellas — inserting domain knowledge while maintaining Nova’s reasoning ability.

Reddit appeared as an early customer example.

AgentCore: Guardrails, Trust & Evaluation

AgentCore received heavy focus — understandably, given AWS’s belief that there will be billions of AI agents in the next era.

A few updates:

AgentCore Policy Controls (New)

“Trust, but verify.”
Automated reasoning systems enforce behavioural and safety boundaries.

AgentCore Evaluations (Preview)

https://aws.amazon.com/blogs/aws/amazon-bedrock-agentcore-adds-quality-evaluations-and-policy-controls-for-deploying-trusted-ai-agents/

Prebuilt evaluation suites let you test agent behaviour across specific criteria, catching issues before they impact customers.

Modernisation at the Speed of AI

The second half of the keynote pivoted to reducing tech debt — something that Matt called a $2.4 trillion problem in the US alone.

AWS made a point with a very literal demonstration on stage (yes, they dropped “tech debt” from a height and exploded it).

AWS Transform (GA)

A full suite covering Windows, mainframe, and custom modernisation with agentic AI.

Kiro Updates

Spec-driven development continues to be a big theme. Kiro now offers:

  • Up to 1 year of Kiro and 100 seats free for qualifying startups
  • Standardisation across internal Amazon teams
  • A new Kiro Autonomous Agent for complex multi-repo changes
  • Security baked into pipelines

AWS Security Agent (Preview)

https://aws.amazon.com/blogs/aws/new-aws-security-agent-secures-applications-proactively-from-design-to-deployment-preview/

AI-driven pen-testing designed to proactively secure applications.

AWS DevOps Agent (Preview)

https://aws.amazon.com/blogs/aws/aws-devops-agent-helps-you-accelerate-incident-response-and-improve-system-reliability-preview/

Incident triage, guided resolution, and fewer 3am alerts.
A big promise — and if it delivers, a big win for platform teams.

The Final 10 Minutes: Rapid Fire

Matt closed with something I’ve never seen before at a re:Invent keynote — 25 announcements in about 10 minutes.

A few highlights:

EC2 Instance Families

macOS also received updates with EC2 M3 Ultra and M4 Max.

AWS Lambda Durable Functions

https://aws.amazon.com/blogs/aws/build-multi-step-applications-and-ai-workflows-with-aws-lambda-durable-functions/

S3 Updates

OpenSearch GPU Vector Acceleration

https://aws.amazon.com/blogs/aws/amazon-opensearch-service-improves-vector-database-performance-and-cost-with-gpu-acceleration-and-auto-optimization/

GuardDuty Extended Threat Detection

https://aws.amazon.com/blogs/aws/amazon-guardduty-adds-extended-threat-detection-for-amazon-ec2-and-amazon-ecs/

Security Hub (GA)

https://aws.amazon.com/blogs/aws/aws-security-hub-now-generally-available-with-near-real-time-analytics-and-risk-prioritization/

CloudWatch Unified Data Store

https://aws.amazon.com/blogs/aws/amazon-cloudwatch-introduces-unified-data-management-and-analytics-for-operations-security-and-compliance/

Database Announcements

Around the Event

After the keynote, it was time to reconnect with partners and walk the Expo — which by Day 2 had gone from a “buzz” to a roar. Easily one of the busiest Expo floors I’ve seen in recent years.

Werner Vogels also dropped his keynote summary and technology predictions for 2026 and beyond. Worth a read: https://www.linkedin.com/pulse/breaking-news-from-aws-reinvent-plus-tech-predictions-2026-beyond-axakc/

A Personal Note

This afternoon I attended the APJC Community Awards and was honoured to receive the Community Award for Ownership.

It’s a meaningful recognition, not because of the trophy, but because of the people behind it — the AWS User Groups, Community Day teams, volunteers, organisers, and friends who put in the real work.

Later in the evening, I attended the Golden Jacket Ceremony, a yearly celebration of everyone who has completed all AWS Certifications. Always a privilege to be part of that moment.

Other Announcements Worth a Look

A massive Day 2 — packed keynote, major launches, personal milestones, and the kind of energy that makes this event what it is.
More tomorrow.