DEV Community

Cover image for AWS re:Invent 2025 – Matt Garman’s Keynote: A Builder’s Breakdown
Adit Modi
Adit Modi

Posted on

AWS re:Invent 2025 – Matt Garman’s Keynote: A Builder’s Breakdown

Day 1 at AWS re:Invent 2025 kicked off with Matt Garman’s first keynote as CEO of AWS — and the message was unmistakable:

This is the AI era, and AWS is building the full stack for it.

Matt structured the entire keynote around four core areas:

  1. AI Infrastructure
  2. Inference Platform
  3. Your Data
  4. Tools to Build Agents

And what I loved most: after each major section, AWS brought customers on stage to share real implementations — not just theory.

Sony, Adobe, and Writer all walked through how they’re building on top of these new capabilities, grounding the announcements in reality.

Here’s my full breakdown.


1. AI Infrastructure: The Foundation for What Comes Next

Matt opened with a thoughtful shoutout to AWS Heroes, Community Builders, and User Group Leaders, calling them the builders shaping the future. As someone involved deeply in community, that meant a lot.

Then came the infrastructure announcements — big ones.

🔹 AWS AI Factories

A new end-to-end framework to scale AI development and deployment. Think of it as the production line for enterprise-grade AI: data → training → eval → deploy.

🔹 Trn3 Ultra Server + the tease of Trn4

AWS rarely brings hardware on stage… but today they did.

Seeing the Trn3 Ultra server physically on the keynote floor made a statement about how seriously AWS is doubling down on custom silicon.

Trn4 previews promise even bigger gains for large-scale training.

(No customer segment here — this section was pure infrastructure.)


2. Inference Platform: Bedrock Gets Even Bigger

Inference is where most real-world workloads live, so this segment was packed.

🔹 New models coming to Amazon Bedrock:

  • Gemma
  • Qwen
  • NVIDIA models
  • Mistral
  • And several domain-specialized models

AWS is clearly pushing for breadth and choice.

🔹 Nova 2 – Sonic, Lite, Pro + Nova 2 Omni

The next-gen Nova models bring stronger multimodal reasoning, faster inference, and improved efficiency.

(This section didn’t have customer stories either — announcements only.)


3. Your Data: Introducing Nova Forge — With Sony On Stage

This was one of the most quietly important announcements, and it got even better when Sony took the stage to show how they’re using it.

🔹 Nova Forge

A new way to standardize, transform, and prepare data specifically for AI workflows — which solves one of the biggest bottlenecks for enterprise teams.

🔹 Sony’s Use Case

Sony showcased how they’re using Nova Forge to power large-scale media, entertainment, and interactive experiences.

They highlighted how data consistency and transformation pipelines directly influence model quality — a real, practical viewpoint that grounded the announcement.

This was one of the most relatable customer segments of the keynote.


4. Tools to Build Agents: Adobe & Writer Take the Stage

This was arguably the most future-facing part of the keynote — AWS clearly sees agents as the new way apps will be built.

🔹 AgentCore Updates

Two major capabilities were introduced:

  • Policy in AgentCore — fine-grained control and governance for agent behavior
  • Evaluations in AgentCore — structured ways to measure, test, and improve agent performance

These are foundational pieces for scaling agent-based applications responsibly.

🔹 New Frontier Agents

AWS previewed a lineup of advanced agents:

  • Kiro — autonomous decision-making
  • Security Agent — automated detection + remediation
  • DevOps Agent — supports deployments, debugging, and operations

🔹 Adobe’s Use Case

Adobe demonstrated how they’re leveraging agent tooling to build more powerful creative workflows — with governance and responsible AI at the center.

🔹 Writer’s Use Case

Writer showed how they’re using agents to create enterprise-grade copilots — emphasizing repeatability, safety, and domain-specific knowledge.

Seeing both Adobe and Writer focus on agent-based systems reinforced AWS’s belief that agents will be the next layer of application logic.


Modernizing the Old: Amazon Transform Custom

AI wasn’t the only focus.

Matt also announced Amazon Transform Custom, a tool built to:

  • Modernize legacy code
  • Upgrade versions
  • Refactor old systems
  • Accelerate modernization projects

This solves problems every enterprise has but rarely talks about publicly.


And in the Last 10 Minutes… All the Non-AI Announcements

Just when everyone assumed the keynote was 100% AI, Matt switched gears and rapid-fired a long list of non-AI updates:

  • Lambda Durable Functions
  • S3 bucket size increases
  • S3 Intelligent Tiering for vector workloads
  • S3 Access Point for NetApp ONTAP
  • Security Hub updates
  • GuardDuty enhancements
  • CloudWatch Unified Data Store
  • And the crowd favorite: Database Savings Plans

A good reminder that AWS is still building across the entire cloud stack — not just AI.


Final Thoughts

This keynote was crisp, focused, and surprisingly grounded in customer reality.

AWS isn’t just shipping models.
They’re building silicon → infra → data → agents → use cases — end to end.

Sony, Adobe, and Writer brought those layers to life with practical examples.

If you want a real-time breakdown, I posted a full live Twitter thread with highlights, reactions, and mini-explanations.

👉 Twitter Thread link here

More updates coming throughout the week. 💙🔥

Top comments (1)

Collapse
 
haerethagomes profile image
haerethagomes

Love this breakdown. Wild stat: AWS says over 50% of new workloads customers discuss with them now have an AI/ML component, which really explains why they're tying silicon, data platforms, and agents together as a single “AI stack.”