Cloud Architect

James Hamilton Portfolio

AWS distinguished engineer whose "Perspectives" blog publishes detailed technical writeups on datacenter network design, custom silicon (Graviton), and the infrastructure engineering decisions behind running cloud services at planetary scale.

AWSDatacenter InfrastructureCloud HardwareSystems Architecture

Location

Remote-first, in the most literal sense — no idea where.

Skills

AWSDatacenter InfrastructureCloud HardwareSystems Architecture

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

Hardware-level detail most cloud architects never publish

Writing about custom silicon design and datacenter network topology goes a level below typical "architecture" content, which usually stops at service design — it shows the infrastructure reasoning underneath the services architects actually use.

2

Historical framing tied to real service milestones

Pieces marking milestones like "EC2's 15-year history" connect specific engineering decisions to a known public timeline, letting readers cross-reference the claims against publicly documented AWS history.

3

A stated engineering philosophy ("constraint-driven innovation")

Naming the recurring theme across posts gives scattered technical writeups a coherent throughline, turning years of individual posts into a single, identifiable point of view on infrastructure design.

What Cloud Architects can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • If you have visibility into a lower layer of the stack than most people in your field discuss, write about that layer specifically — it differentiates you from generic architecture commentary.

  • Tie technical writeups to real, checkable milestones or dates so readers can verify the claims against public history.

  • Name the recurring theme or philosophy behind your work explicitly — it turns a scattered archive into a single, memorable point of view.

Case studies

Perspectives goes below the service layer: a flat datacenter network that uses 69% fewer routers, AWS's custom-silicon history, and a keynote on how hardware constraints shape databases. James Hamilton's posts cite papers and give real numbers.

Case StudyadvancedPublished June 2026

Flat Datacenter Networks at Scale

Datacenter networkingRandom graph / expander theoryFat-tree (Clos) topologiesValiant Load Balancing

A writeup of Amazon's Resilient Network Graphs (RNG), which replace hierarchical fat-tree datacenter networks with flat, quasi-random topologies. It traces the history from 1970s expander graphs and his own 2009 VL2 paper, through Jellyfish, to how an AWS team solved routing (Spraypoint), cabling (the ShuffleBox) and operations.

It explains a production redesign from theory through deployment, including the limitations. That kind of first-principles reasoning is rare in cloud writing.

  • 69% fewer routers, 33% more throughput, 40% less network power vs fat-trees

  • First RNG data center near Dublin in 2024, then Germany and Spain

  • Builds on VL2 (SIGCOMM 2009), which Hamilton co-authored

Case StudyadvancedPublished August 27, 2022

A Short History of AWS Silicon Innovation

Custom ASICsAWS NitroGravitonSemiconductor design

A post and video on why a cloud provider designs its own chips. It goes back to 2012, introduces the key people, and walks through the early decisions and how AWS's semiconductor investment grew into Nitro and Graviton. In the comments, Hamilton confirms AWS remains fabless and partners with TSMC.

It connects hardware investment to cloud economics and the services built on it. That context is behind choices like Graviton instances.

  • 10-year history starting in 2012

  • 20M+ AWS-designed ASICs deployed, per the post

  • Author answers reader questions in the comments (fabless, TSMC)

Projects

ResearchadvancedDelivered January 2024

Constraint Driven Innovation (CIDR 2024 keynote)

Database systemsHardware accelerationWrite-ahead loggingIn-memory databases

The opening keynote at the Conference on Innovative Data Systems Research 2024. The thesis is that constraints both force and block innovation: slow disks produced write-ahead logging, while in-memory databases waited decades for cheap large-memory servers. He argues the database market is now large enough to justify custom hardware.

It reads hardware cost curves as a guide to where software architecture is heading, which is useful for long-range platform bets.

  • Kicked off CIDR 2024 with the keynote

  • Examples: WAL from slow disks; in-memory DBs waiting on RAM costs

  • Makes the case for custom database hardware

Ready to build your portfolio?

Follow the Cloud Architect roadmap — skills, projects, and timeline to get hired.

Cloud Architect Roadmap