Amazon SQS at 20: How a Simple Queue Became the Backbone of Cloud Infrastructure

On July 13, 2006, Amazon launched three cloud services that would define modern infrastructure: EC2, S3, and SQS. Two decades later, the first two get most of the attention. SQS may be the one that actually kept everything running.

What Happened

Amazon Simple Queue Service launched as a managed way to pass messages between distributed components without requiring them to be available at the same time. The original 8 KB message limit was narrow even by 2006 standards, but the use case was durable: decouple services so that a spike or failure in one doesn’t cascade to another.

Twenty years later, the payload limit is 1 MiB (with S3-backed payloads reaching 2 GB), FIFO queues support 70,000 transactions per second in high-throughput mode, and JSON protocol support in the SDK has reduced end-to-end latency by up to 23% for typical message sizes.

The milestones that matter in practice: dead-letter queue redrive moved to SDK/CLI in 2023, in-flight FIFO message limits jumped from 20,000 to 120,000 in 2024, and fair queuing for multi-tenant workloads arrived in 2025. Each reflects the same pattern — SQS becoming less of a simple buffer and more of a workload-aware routing layer.

Why It Matters

The throughput trajectory tells the real story. High-throughput FIFO mode scaled from 3,000 TPS at launch in 2021 to 70,000 TPS by November 2023 — a 23x increase in two years. That isn’t incremental improvement; it’s a category shift toward serving AI inference pipelines, high-volume event streams, and multi-tenant SaaS workloads that didn’t exist when SQS launched.

The fair queuing addition in 2025 is the most architecturally significant recent change. Traditional SQS FIFO serves messages in strict order, but in multi-tenant environments one noisy consumer can starve others. Fair queuing distributes processing capacity across queue groups — a direct response to the reality that modern SQS deployments serve dozens of parallel workflows, not a single service-to-service link.

What This Means for Builders

Two decades of SQS evolution points at a consistent principle: the value isn’t in the queue itself, it’s in what decoupling lets you do elsewhere. Every throughput increase, every payload expansion, every FIFO improvement has extended the scenarios where you can say “add SQS here” instead of building direct synchronous coupling.

For teams currently designing AI agent pipelines, SQS’s 2025-era feature set — fair queuing, 120,000 in-flight FIFO messages, 1 MiB native payloads — maps well to the task dispatch patterns agentic workloads need. The primitives that power the web are starting to look like the right primitives for agent orchestration, too.

Source: AWS News Blog

Content created with AI assistance and reviewed for accuracy.