re:Invent 2021: AWS and the $71 Billion Run Rate

I just got back from Las Vegas. AWS re:Invent 2021 ran from November 29 through December 3, and after two years of virtual events, being there in person felt like a homecoming for the cloud community. Tens of thousands of attendees, hundreds of sessions, and the usual firehose of product announcements that takes weeks to fully digest.

But the number that sticks with me is $71 billion. That is AWS's annualized revenue run rate, announced during Adam Selipsky's keynote. To put that in perspective, AWS alone generates more revenue than most Fortune 100 companies. A business unit that started as an internal infrastructure project sixteen years ago is now one of the most profitable enterprises on the planet.

As someone who builds on AWS every day at a major entertainment company, the scale of this ecosystem is both exciting and sobering.

The Keynote Highlights

Adam Selipsky, who returned to AWS as CEO earlier this year after a stint running Tableau, delivered a keynote that was more polished and narratively coherent than Andy Jassy's previous keynotes. Where Jassy was exhaustive, listing every feature and service with engineering precision, Selipsky organized the announcements around customer stories and industry themes.

The major announcements that matter most for my work:

AWS Graviton3 processors were announced, the third generation of AWS's custom ARM-based chips. Graviton2 instances already deliver better price-performance than comparable Intel-based instances for many workloads. If Graviton3 delivers the claimed 25% improvement over Graviton2, the case for migrating workloads to ARM becomes increasingly difficult to ignore. We have been running some workloads on Graviton2 at the company, and the results have been compelling enough that Graviton3 will be an immediate evaluation priority.

Amazon SageMaker Canvas brings no-code machine learning to business analysts. This is part of a broader trend of making ML accessible beyond data scientists, and while I am skeptical of no-code approaches for serious ML work, the democratization of basic predictive analytics is worth watching.

AWS Lake Formation governed tables address one of the hardest problems in data engineering: managing transactional consistency across data lakes. ACID transactions on S3-backed data is something my colleagues in data engineering have been asking about for years.

Amazon Inspector got a complete rebuild. The new version provides continuous, automated vulnerability scanning for EC2 instances and container images with findings consolidated in a single view. Security scanning that works automatically without manual configuration is exactly the kind of operational improvement that matters at scale.

The Cloud Market in Numbers

The $71 billion run rate deserves more analysis because the growth trajectory is remarkable even by tech industry standards.

AWS launched S3 in 2006. It took about ten years to reach a $10 billion run rate. It took another three years to reach $30 billion. And it has roughly doubled again in the two years since. The growth is accelerating in absolute terms even as the base gets larger, which defies the typical pattern where growth rates slow as businesses mature.

The pandemic accelerated cloud adoption across every industry. Companies that had been debating multi-year migration timelines suddenly needed to support remote workforces, scale digital channels, and build resilience into systems that had been running on aging data center hardware. The result was a surge in cloud spending that benefited all three major providers but AWS most of all, given its first-mover advantage and breadth of services.

Azure is growing faster in percentage terms, and Google Cloud is investing aggressively, but in absolute revenue, AWS remains the clear leader. The competitive dynamics are fascinating to watch from the inside. We evaluate Azure and GCP for specific workloads, and the gap in service maturity that was enormous five years ago has narrowed considerably. But the ecosystem effects, the breadth of services, the partner network, the training and certification infrastructure, the depth of documentation, all favor AWS in ways that are hard to replicate.

What I Saw on the Ground

Beyond the keynotes, re:Invent is valuable for the conversations that happen in hallways, at meals, and during the expo hall wanderings. The themes I heard repeatedly from other enterprise architects:

Multi-cloud is real but messy. Almost every enterprise I spoke with runs workloads on more than one cloud provider, but the dream of seamless portability between clouds remains just that, a dream. In practice, organizations choose a primary cloud and use secondary providers for specific workloads or as a negotiating lever. The operational overhead of truly multi-cloud infrastructure, with consistent security, monitoring, and deployment practices across providers, is higher than most teams anticipate.

Kubernetes is the new default. EKS adoption is growing rapidly, and the conversations have shifted from "should we use Kubernetes" to "how do we operate Kubernetes at scale." The tooling ecosystem around EKS, including add-on management, observability, and service mesh, was a major focus of breakout sessions.

Cost management is the new crisis. Multiple people told me variations of the same story: cloud spending grew 40-60% year over year during the pandemic, and now finance teams are asking hard questions. The ease of provisioning resources in the cloud, which is a feature, also makes it easy to overspend, which is a problem. FinOps as a discipline is moving from nice-to-have to essential.

Security is shifting left, finally. The integration of security scanning into CI/CD pipelines, the automation of compliance checks, and the use of infrastructure-as-code policy frameworks like OPA and Sentinel were common topics. Security is no longer something that happens after deployment; it is being embedded into the development process.

What This Means for Enterprise Architecture

For those of us building cloud platforms at scale, re:Invent 2021 reinforced several trends that were already underway.

The platform engineering model is winning. Organizations that build internal developer platforms, abstracting away the complexity of raw cloud services behind opinionated, self-service interfaces, are moving faster and spending less than organizations that give every team direct access to the cloud console. The challenge is building these platforms without recreating the rigidity of the on-premises world they replaced.

Serverless is expanding beyond Lambda. The serverless model, where you write code and the cloud handles everything else, is extending to databases (Aurora Serverless), containers (Fargate), and analytics (Athena, Redshift Serverless). The operational benefits of not managing infrastructure are significant, and the pricing models are increasingly competitive with provisioned alternatives.

The talent gap is the real bottleneck. Every conversation I had about ambitious cloud strategies eventually arrived at the same constraint: "we cannot hire enough people who know how to do this." The demand for cloud engineers, platform architects, and SREs far exceeds the supply, and the gap is widening as adoption accelerates.

Looking Ahead

AWS at $71 billion is not slowing down. The investment in custom silicon (Graviton, Trainium, Inferentia) signals a long-term strategy to differentiate on hardware performance while maintaining the service abstraction that customers expect. The expansion into industry-specific solutions (healthcare, financial services, automotive) suggests a move up the value chain from infrastructure to industry platforms.

For my work, the priorities coming out of re:Invent are clear: evaluate Graviton3 for our compute workloads, expand our EKS footprint with the new operational tooling, implement the rebuilt Inspector across our account fleet, and continue building the internal developer platform that makes all of this accessible to application teams without requiring them to become cloud experts.

Another re:Invent in the books. Time to process the announcements, update the roadmap, and get back to building.