Outposts

Some workloads must stay physically on-premises — for single-digit-millisecond latency to factory equipment or trading systems, data-residency rules, or local processing that can’t depend on a WAN link — yet the team wants the AWS control plane, APIs, and services, not a separate on-prem stack. AWS Outposts delivers that: AWS ships, installs, and fully manages its own hardware (a 42U rack, or smaller 1U/2U servers) in your facility. It’s an extension of an AWS Region: you launch EC2, EBS, ECS, EKS, S3 on Outposts, RDS, and more on the local capacity through the normal console/API/CloudFormation.

• An Outpost is anchored to one AWS Region (its “home Region”) and appears in your account like an extension of one AZ. It is not multi-AZ on its own — for local HA you deploy multiple Outposts.
• It needs a reliable Service Link (network path) back to the home Region for control-plane operations, monitoring, and updates; the local gateway carries data-plane traffic to your on-prem network.
• If the link to the Region drops, already-running instances keep running, but you can’t launch new ones or make API changes until it recovers.

graph LR
subgraph REGION["AWS Region (home)"]
  CP["AWS control plane / console / APIs"]
end
CP <-->|"Service Link (VPN over your network)"| OP
subgraph DC["Your Data Center"]
  subgraph OP["AWS Outpost (AWS-owned, AWS-managed)"]
    EC2["EC2 + EBS"]
    CON["ECS / EKS"]
    S3O["S3 on Outposts"]
    RDSO["RDS on Outposts"]
  end
  LGW["Local Gateway"] --> ONPREM["On-prem apps / OT equipment"]
  OP --> LGW
end
classDef op fill:#f3ecfb,stroke:#8b3de0,stroke-width:2px,color:#3a1a5c
classDef region fill:#eaf7ec,stroke:#2e8b3d,stroke-width:2px,color:#1a3d20
class EC2,CON,S3O,RDSO op
class CP region

You pay for the Outpost capacity up front or monthly over a 3-year term (all-upfront, partial, or no-upfront), covering the hardware, delivery, installation, and maintenance — you’re effectively reserving a fixed pool of compute and storage. Services that reach back to the Region (data transfer to/from the home Region, S3 in-Region, etc.) bill normally. There’s no per-instance On-Demand rate for capacity that’s already on the Outpost — you sized and paid for it.

• Trigger phrases: “must run on-premises,” “very low latency to local systems,” “data residency / cannot leave the building,” “but wants native AWS services and APIs” → Outposts.
• Outposts vs Local Zones vs Wavelength: Local Zones and Wavelength are AWS-operated sites near cities / in telco 5G networks — you don’t host anything. Outposts is hardware in your own facility.
• Outposts vs Snow Family: Snow is for temporary / disconnected / edge / data-transfer use and ships back; Outposts is a permanent, connected extension of a Region.
• Not inherently highly available — one Outpost is like part of one AZ. Design HA with multiple Outposts or by failing back to the Region.
• Supports EC2, EBS, ECS, EKS, S3 on Outposts, RDS, ElastiCache, EMR, and Application Load Balancer locally; the exact service list is smaller than a full Region.
• Fully AWS-owned and AWS-maintained — your staff never patches or replaces the hardware.