S3 Express One Zone
An AWS S3 storage class delivering single-digit millisecond latency for frequently accessed data, using Directory Buckets in a single Availability Zone. Scales to 200,000 PUT and 2,000,000 GET TPS per bucket and charges 50% less per request than S3 Standard. An 85% storage price reduction in early 2025 transformed the TCO equation.
Summary
An AWS S3 storage class delivering single-digit millisecond latency for frequently accessed data, using Directory Buckets in a single Availability Zone. Scales to 200,000 PUT and 2,000,000 GET TPS per bucket and charges 50% less per request than S3 Standard. An 85% storage price reduction in early 2025 transformed the TCO equation.
S3 Express One Zone fills the performance gap between standard S3 and local attached storage. It enables latency-sensitive workloads — ML training data loading, interactive analytics, real-time feature serving, agent scratchpad state — to use S3 without the cold-start penalty. Lyrebird Studio publicly reported an 18% overall TCO reduction (80% faster workflow operations, 11% lower compute provisioning) after moving intermediate generative-AI state into Express One Zone.
- Not multi-AZ. Data resides in a single AZ; not a replacement for S3 Standard for durable primary storage.
- Zonal endpoints demand AZ affinity. Cross-AZ compute access negates the performance gain — PUT tail latencies stretch into seconds.
- Idle buckets (no requests for 90 days) auto-transition to an inactive state that returns HTTP 503 until reactivated, even though storage charges keep accruing. Budget for that in infrequently-touched workspaces.
implementsS3 API — same API, directory bucket semanticsdepends_onS3 Directory Bucket — the underlying namespace constructsolvesCold Scan Latency — single-digit ms first-byte accesssolvesHigh Cloud Inference Cost — reclaims GPU cycles lost to I/O wait in agent pipelinesscoped_toDirectory Buckets / Hot Object Storage — AWS implementation of this concept
Definition
AWS's purpose-built low-latency storage class using **S3 Directory Bucket** topology within a single AZ, delivering up to 10x faster performance than S3 Standard with single-digit millisecond first-byte latency. An 85% price reduction in early 2025 repositioned it as a viable high-performance buffer tier for streaming and analytics pipelines. Per-bucket capacity scales to **200,000 PUT / 2,000,000 GET TPS** — roughly 60× the partitioned-prefix throughput of General Purpose buckets — and request pricing is 50% lower than standard tiers.
Standard S3 is optimized for durability across 3 AZs, but this adds latency. Express One Zone trades multi-AZ redundancy for speed, targeting AI/ML training, real-time analytics, and high-frequency trading. The 2025 price cut made it economical for systems like WarpStream to use Express One Zone as a high-throughput buffer before compacting data into S3 Standard. For agentic AI workflows, the low-latency scratchpad collapses the I/O waits that keep GPUs idle between inference steps — Lyrebird Studio reported **80% faster** sequential workflow operations, **11%** lower compute provisioning, and **18%** overall TCO reduction after migrating intermediate state to Express One Zone.
AI/ML training data staging, real-time analytics acceleration, high-frequency checkpoint storage, interactive query acceleration, low-latency buffer tier for streaming pipelines, agentic AI scratchpad memory and intermediate state sharing.
Connections 10
Outbound 8
implements1depends_on1constrained_by1Inbound 2
depends_on1enables1Resources 5
Official user guide for S3 Express One Zone directory buckets covering creation, access, and performance characteristics.
Product page with pricing, feature overview, and use-case guidance for the single-digit-millisecond storage class.
Launch announcement with architecture details and performance benchmarks showing up to 10x faster request processing.
Design patterns for hitting the 200K PUT / 2M GET TPS ceiling — AZ affinity, request distribution, client concurrency.
Published case study with measured outcomes — 80% faster workflow operations, 11% compute reduction, 18% overall TCO cut — for generative AI workloads migrating to Express One Zone.