DEV Community

Cover image for My Takeaways from the Peter DeSantis & Dave Brown Keynote — re:Invent 2025
Adit Modi
Adit Modi

Posted on

My Takeaways from the Peter DeSantis & Dave Brown Keynote — re:Invent 2025

I’ll admit it upfront: I didn’t make it to the keynote hall for this one.
Jetlag + late night writing = I woke up late and had to watch the keynote afterward.
But honestly? Even on replay, this keynote delivered exactly what I expect from Peter and Dave every year — deep tech, real numbers, and a behind-the-scenes look at how AWS is engineered.

This keynote isn’t about AI announcements or product hype.
It’s about the infrastructure, silicon, and systems thinking that make everything on AWS possible.

The Five Unchanging Pillars of AWS

Peter opened with a reminder of the core attributes AWS has optimised around for nearly two decades:

Security

Availability

Elasticity

Cost

Agility

Even in a rapidly evolving AI world, these don’t change — only the techniques to achieve them do.

It was a good grounding moment. Before we talk about the future of AI, we need to understand the foundation that makes any of it possible.

Nitro — The Engine Behind AWS Compute

Peter used the Nitro System as the perfect example of AWS’s deep, multi-year investment strategy.

Nitro continues to be:

Custom silicon

Minimal overhead

Isolation by design

Virtually no “jitter”

And now… literally part of CS textbooks

Nitro isn’t a feature. It’s AWS’s DNA.

Graviton — Cloud CPUs Done Right

Dave Brown came on stage next, and he delivered absolute gold.

Highlights:

Graviton keeps delivering the best price-performance for cloud workloads.

Real-world customer impact from Adobe, Epic Games, Formula 1, Pinterest, SAP.

Direct-to-silicon cooling for better sustainability & thermal efficiency.

Graviton 4 doubling L2 cache.

And then the big one: Graviton 5

192 cores

More L3 cache

Up to 25% better performance

AWS isn’t slowing down. They’re accelerating.

Apple on Graviton — A Rare and Powerful Guest Moment

One of the most interesting segments was Payam Mirrashidi from Apple.

He shared:

How Apple uses Swift on Graviton for huge-scale services (App Store, Music, TV).

Why Swift’s performance & safety make it ideal for modern cloud apps.

How homomorphic encryption + Swift power privacy-preserving features like spam detection.

And the announcement: native Swift toolchain for Amazon Linux.

That’s a big shift for Swift developers and cloud-native teams.

Lambda’s Evolution + Lambda Managed Instances

Peter returned to talk serverless, tracing the journey from the birth of Lambda in 2014 to where we are now.

And then came something interesting:

Lambda Managed Instances

A new way to blend:

Serverless simplicity

EC2-like control

Customers can pick instance types and hardware, while Lambda still manages scaling, patching, and provisioning.

This is AWS listening to real customer pain points — especially those needing predictable performance with serverless ergonomics.

Project Mantle — Rebuilding Inference for the AI Era

This might be one of the most important (but underrated) announcements.

Peter introduced Project Mantle, the new inference engine that powers Amazon Bedrock.

Key capabilities:

Service tiers for different latency/urgency needs

Customer-specific queues to guarantee fairness

Adaptive capacity sharing

A durable journal for fault tolerance

Confidential computing for high-security workloads

Inference is becoming the real bottleneck in AI — and Mantle feels like AWS’s long-term answer.

Vector Search Everywhere

Peter then shifted to vector search, breaking it down in a way that actually makes sense:

Embedding models convert data into multi-dimensional vectors

Vector databases find nearest neighbors

Suddenly, unstructured knowledge becomes searchable

This isn’t just for search engines — it’s for any org trying to unlock institutional knowledge across docs, media, chats, and more.

Agentic Development + AWS Agent

This was a smooth transition into agentic development.

Peter explained how agents become a new abstraction layer —
one that uses your existing systems, APIs, and knowledge bases,
but automates multi-step tasks end-to-end.

Then came AWS Agent — the tool to actually build those agents:

Connect to your tools & data

Define goals

Let agents plan & act

Even generate pull requests

Seeing an agent analyze feedback → design a feature → submit a PR was surreal.
This is no longer “future tech.” It’s happening.

Closing Thoughts

Peter ended with a simple message:
AWS solves the hard problems so builders can invent faster.

After watching this keynote, that mission feels more alive than ever — from silicon, to serverless, to inference engines, to agentic automation.

Even though I watched it late (because I definitely overslept), this keynote reminded me why AWS continues to lead:
they build deep, they build long-term, and they build for builders.

Top comments (0)