Free lesson · AWS Fundamentals

Edge locations, Local Zones, Wavelength, and Outposts

Easy7 min read

Beyond Regions And AZs

Regions and AZs are the primary infrastructure model, but they are not the whole map. AWS operates four additional infrastructure tiers, each one trading breadth of service coverage for physical proximity to something - end users, mobile networks, or your own data center.

You reach for these when a Region cannot satisfy a latency, locality, or residency requirement. Not before. Every tier below a Region has fewer services, higher per-unit cost, and more operational complexity than the Region it extends.

The Proximity Ladder

graph LR
    U[End User / Device] --> W[Wavelength Zone<br/>inside 5G network]
    U --> E[Edge Location<br/>CloudFront PoP]
    U --> L[Local Zone<br/>metro area]
    U --> O[Outpost<br/>your data center]
    E --> RE[Regional Edge Cache]
    RE --> R[Parent Region]
    L --> R
    W --> R
    O --> R
    R --> AZ[Availability Zones]

This diagram shows the AWS proximity ladder as a flow from the end user on the left to the Region on the right. An end user can be served by a Wavelength Zone inside a 5G network, an Edge Location which is a CloudFront point of presence, a Local Zone in their metro area, or an Outpost in a private data center. Edge Locations fall back to Regional Edge Caches, which fall back to the parent Region. Local Zones, Wavelength Zones, and Outposts all depend on a parent Region, which contains the Availability Zones.

The critical structural fact: every tier below the Region is an extension of a parent Region, not a replacement for it. Control plane operations, identity, and most managed services still resolve back to the parent Region. If the link to the parent Region degrades, the extension degrades with it.

Edge Locations

The largest tier by count - hundreds of points of presence worldwide, far more than there are Regions.

What runs there:

Regional Edge Caches sit between PoPs and your origin as a second cache tier. A PoP miss checks the Regional Edge Cache before going to origin, which meaningfully improves hit ratios for large or less-popular objects.

Tip: Edge locations are not compute infrastructure in any general sense. You cannot run a container or attach a volume there. CloudFront Functions and Lambda@Edge exist to manipulate requests and responses - header rewriting, redirects, auth checks, A/B routing - not to host application logic. If you find yourself wanting a database at the edge, you want a different tier or a different design.

Local Zones

An extension of a Region placed in a metropolitan area far from the parent Region's AZs. Purpose: single-digit millisecond latency to users or systems in that metro when the parent Region is hundreds of miles away.

Characteristics:

• Presented as an additional zone in the parent Region. You extend an existing VPC into it with a subnet.
• A subset of services: typically EC2, EBS, VPC networking, Application Load Balancer, and a limited set of managed services. Not the full catalog.
• Most Local Zones are effectively a single fault domain - there is no "three Local Zones" redundancy pattern within one metro.
• Priced at a premium relative to the parent Region.

When it is the right answer: real-time media production, interactive rendering, gaming, industrial systems, and financial workloads where the user population is concentrated in one metro and 30ms is too slow.

When it is not: general web applications. CloudFront in front of a Regional origin solves the latency problem for read-heavy workloads at a fraction of the cost and complexity.

Wavelength Zones

AWS infrastructure deployed inside a telecommunications provider's 5G network, at the network edge. Traffic from a mobile device reaches your compute without traversing the public internet or the provider's core network.

• Target latency: ~10ms or better from device to application.
• Use cases: AR/VR, connected vehicles, real-time video analytics on mobile, industrial IoT over cellular.
• Constraints: reachable primarily from devices on that carrier's network in that geography. Availability is tied to specific carrier partnerships.

Wavelength is a narrow tool. If your users are not on mobile networks, and the latency budget is not sub-20ms, it is almost certainly not the answer.

Outposts

AWS-designed and AWS-managed physical hardware installed in your data center or colocation facility, running AWS services locally and connected back to a parent Region.

Two form factors: full 42U racks, and smaller servers for space-constrained locations such as retail floors or factories.

Why anyone chooses it:

• Data residency where no Region exists in the required jurisdiction
• Local processing latency to on-premises systems - a factory floor, a trading system, a hospital imaging network
• Migration bridge for applications that cannot yet leave the data center but should use cloud tooling

What it costs you:

• You buy or commit to capacity, typically on multi-year terms. This is the opposite of elastic.
• AWS owns maintenance, but you own power, cooling, space, and physical security.
• Service coverage is a subset - EC2, EBS, S3 on Outposts, ECS/EKS, RDS on Outposts, and a limited set of others.

Warning: The parent Region dependency is the failure mode people underestimate. An Outpost that loses its network link to the parent Region enters a degraded state: existing instances keep running, but control-plane operations - launching instances, attaching volumes, most API calls, and in many configurations IAM authentication - become unavailable. Outposts is not a disconnected-operation product. If your requirement is "must keep working with no connectivity to AWS," Outposts alone does not meet it.

Choosing A Tier

Pros:
- Edge locations: cheapest and simplest latency win, no infrastructure to manage
- Local Zones: real single-digit-millisecond compute in a metro, uses normal VPC constructs
- Wavelength: unmatched latency for mobile-network users
- Outposts: satisfies residency and on-premises latency requirements nothing else can

Cons:
- Edge locations: cache and request manipulation only, no general compute
- Local Zones: single fault domain, partial service catalog, premium pricing
- Wavelength: carrier and geography locked, very narrow use case
- Outposts: capital commitment, physical operations burden, degrades without the parent Region

Important: Work down this list, not up. The overwhelming majority of latency problems presented as "we need edge compute" are solved by CloudFront with correct cache headers, a Regional deployment closer to the users, or fixing an N+1 query. Reach for Local Zones, Wavelength, or Outposts only after you have measured where the milliseconds actually go.

Cost Shape

The shape matters more than the rate. Edge is fully elastic. Outposts is the least elastic thing AWS sells - you are back to capacity planning, with a multi-year commitment and no ability to scale past what you installed.

Why This Matters In An Interview

Interview: What interviewers are listening for:

• You default to CloudFront for latency and only escalate tiers when the requirement genuinely demands it. Reaching for Outposts in a general system design question reads as pattern-matching, not reasoning.
• You know that these tiers extend a parent Region rather than standing alone, and you can name what breaks when the link degrades.
• You can distinguish CloudFront Functions from Lambda@Edge - a small detail that reliably separates people who have used the edge from people who have read about it.

Common follow-ups:

• "Users in Australia report slow load times. What do you do?" → Measure first: is it TTFB (origin problem), transfer (CDN problem), or DNS? Then CloudFront, then consider a Regional deployment in ap-southeast-2, then consider whether the write path needs to be regional too.
• "Regulator says data cannot leave the country and there is no Region there. Options?" → Outposts, or a Local Zone if one exists in that jurisdiction, or a partner/sovereign cloud arrangement. Then discuss the operational cost honestly.
• "Why not just put everything at the edge?" → Because state has to live somewhere, and edge tiers have no consistent, durable, transactional storage. The edge is good at read-mostly and request manipulation.

Practice

Exercise: Starter. Look up which Local Zones and Wavelength Zones exist near you today, and note which services each supports. Compare the service list against a full Region.

Intermediate. Take an application you know and identify every millisecond in the user-perceived latency budget: DNS, TCP, TLS, CDN lookup, origin round trip, application processing, database. Decide which tier - if any - would actually help.

Advanced. Design a connected-vehicle telemetry system with a 20ms budget for safety-critical responses and a relaxed budget for analytics. Split the workload across tiers and justify each placement, including what happens when the vehicle's carrier connection degrades.

Related Topics

• Module 03 - Networking: CloudFront, Global Accelerator, Route 53 routing policies
• Module 04 - Compute: Lambda execution model, and why Lambda@Edge is constrained
• Module 14 - Migration & Hybrid: Outposts, ECS Anywhere, EKS Anywhere

Back to AWS Fundamentals