Amazon Web Services (AWS) Performance: Benchmarks, Latency & Limits 2026
AWS App Runner bills $0.064/vCPU-hour and scales on concurrency. Lightsail runs $5 to $1,764/month, and EC2 carries a 99.99% regional SLA.
Amazon Web Services (AWS) Performance verdict
AWS hosts apps two ways.
App Runner runs managed containers at $0.064 per vCPU-hour plus $0.007 per GB-hour, scaling on concurrency. Lightsail sells fixed instances from $5 to $1,764 a month.
Start with the timing. For new managed-container workloads, use ECS Express Mode rather than App Runner, which no longer takes new customers. Keep at least one provisioned instance warm to avoid cold starts, and budget by active vCPU-seconds. For fixed, predictable cost, Lightsail fits. For raw scale and the full region footprint, drop to EC2. Pin compute to the region nearest your users and their data.
- App Runner stopped accepting new customers on April 30, 2026. New projects should use Amazon ECS Express Mode, which keeps the same usage-based pricing.
- There is no dedicated App Runner or Lightsail SLA page. The EC2 compute SLA, 99.99% regional and 99.5% per instance, is the closest guarantee.
- App Runner scales on concurrency, but the docs prose never states the default value. Pricing also runs higher in Asia Pacific, Tokyo especially.
- Regions / AZs
- 39 / 123
- CloudFront PoPs
- 750+
- EC2 SLA (region)
- 99.99%
- App Runner rate
- $0.064/vCPU-hr
- App Runner status
- Closed to new
This page covers how AWS compute performs and scales, and what it costs. Region coverage lives on its own page.
Estimate your Amazon Web Services (AWS) usage cost
- 1 × (1 vCPU + 1 GB) for 730 h/mo costs about $51.86/mo (90% CPU, 10% RAM), running always-on.
- App Runner stopped accepting new customers on April 30, 2026; ECS Express Mode is the successor with the same usage-based model. Tokyo rates are higher.
Estimated from Amazon Web Services (AWS)'s published per-minute rates ($0.001067/vCPU-min, $0.000117/GB-min). Egress and storage are billed separately. Verify against your own workload.
Reliability and network signals
| Signal | Value | Source |
|---|---|---|
| EC2 SLA (region) | 99.99% | AWS Compute SLA |
| EC2 SLA (instance) | 99.5% | AWS Compute SLA |
| CloudFront PoPs | 750+ / 100+ cities | CloudFront |
| Embedded PoPs | 1,140+ / 300+ cities | CloudFront |
| Global backbone | 20M+ km fiber | AWS Global Infrastructure |
| App Runner warm latency | Low-ms (provisioned) | App Runner |
| Lightsail-class price (independent) | ~$0.027/hr (2 CPU/4 GB) | Cloud Mercato |
Managed-container scaling
| Capability | Value | Notes |
|---|---|---|
| Scaling metric | Concurrency | Max simultaneous requests per instance before scaling up |
| Min / max size | Configurable | Min provisioned instances kept; max scale-up ceiling |
| Idle behavior | 1 provisioned (warm) | Falls back to provisioned instances when idle; no cold start |
| Billing | Per second, 1-min vCPU min | Active instances bill vCPU + memory per second |
| Auto scaling configs | 10 / region, 5 revisions | Reusable named configurations |
| Instance ceiling | Fargate vCPU quota | Instance count tied to Fargate On-Demand vCPU quota |
Compute options and prices
| Option | vCPU / RAM | Price |
|---|---|---|
| App Runner (active) | 0.25, 4 vCPU / 0.5, 12 GB | $0.064/vCPU-hr + $0.007/GB-hr |
| App Runner (provisioned) | warm, memory only | $0.007/GB-hr |
| Lightsail | 2 vCPU / 0.5 GB | $5/mo |
| Lightsail | 2 vCPU / 4 GB | $24/mo |
| Lightsail | 4 vCPU / 16 GB | $84/mo |
| Lightsail | 16 vCPU / 64 GB | $384/mo |
| Lightsail (max) | 64 vCPU / 256 GB | $1,764/mo |
Amazon Web Services (AWS) reliability and architecture
- Amazon EC2 carries a 99.99% regional SLA when instances run across two or more Availability Zones, and a 99.5% single-instance SLA, with credits when an instance is unavailable more than six minutes in a clock-hour
- App Runner stopped accepting new customers on April 30, 2026; existing services keep running and AWS recommends Amazon ECS Express Mode as the successor
- App Runner keeps provisioned instances warm while idle so requests get consistently low millisecond latency with no cold start, defaulting to one provisioned instance
- CloudFront fronts applications with 750+ points of presence in 100+ cities, 15 regional edge caches, and 1,140+ embedded PoPs in 300+ cities
- AWS holds ISO 27001/27017/27018/27701, SOC 1/2/3, PCI DSS and CSA STAR globally, plus HIPAA, FedRAMP (via GovRAMP), FISMA and CJIS for US public sector
- AWS operates an independent European Sovereign Cloud located entirely within the EU for digital-sovereignty requirements
Amazon Web Services (AWS) performance benchmarks, independently measured
- Cloud Mercato ran a 13-year performance study of five EC2 general-purpose generations (m3 through m8g) at a fixed 8 vCPU / 32 GB spec, using sysbench CPU and RAM, OpenSSL Speed, Geekbench 6, bandwidth and network-latency tests
- Cloud Mercato's public catalog independently prices Lightsail instances, for example a 2 CPU / 4 GB instance at $0.027/hour and an 8 CPU / 32 GB instance at $0.213/hour
- Cloud Mercato also benchmarks the AWS global network, measuring bandwidth, latency and hops between all AWS regions
- App Runner active instances bill per second, with a one-minute minimum vCPU charge each time a provisioned instance starts processing requests
- App Runner is available in 11 AWS regions, a subset of the broader 39-region footprint
- Asia Pacific (Tokyo) App Runner pricing is higher than US/EU, at $0.081/vCPU-hour and $0.009/GB-hour
Amazon Web Services (AWS) Performance FAQ
How does AWS host applications?
Several ways. App Runner is the managed-container service that builds and deploys a container, then autoscales on concurrency, at $0.064 per vCPU-hour plus $0.007 per GB-hour. It stopped accepting new customers on April 30, 2026, with ECS Express Mode as the successor. Lightsail offers fixed-price instances from $5 to $1,764 a month. EC2 is the core compute, all on the 39-region, 123-AZ footprint with 750+ CloudFront edge points.
How does App Runner autoscaling work?
It scales on concurrency, the maximum number of simultaneous requests a single instance handles. When concurrent requests pass that limit, App Runner adds active instances up to your max size. When traffic drops, it scales back to provisioned warm instances, one by default. Active instances bill per second with a one-minute minimum vCPU charge, and the instance ceiling ties to your Fargate On-Demand vCPU quota.
What does AWS compute cost?
App Runner active instances are $0.064 per vCPU-hour plus $0.007 per GB-hour, with idle provisioned instances billed memory-only at $0.007 per GB-hour, and Tokyo priced higher. Lightsail is fixed monthly, from $5 for 0.5 GB and 2 vCPU up to $1,764 for 256 GB and 64 vCPU, with IPv6-only bundles slightly cheaper. EC2 spans on-demand, savings plans and spot, benchmarked independently by Cloud Mercato.
What SLA and reliability does AWS provide?
EC2 carries a 99.99% regional uptime SLA when instances run across two or more Availability Zones, plus a 99.5% single-instance SLA. Service credits kick in if an instance is down more than six minutes in a clock-hour. There is no separate App Runner or Lightsail SLA page, so the compute SLA applies. App Runner keeps provisioned instances warm to avoid cold starts.
Is AWS independently benchmarked?
Yes. Cloud Mercato, an independent benchmark lab, ran a 13-year study of five EC2 generations using sysbench, OpenSSL, Geekbench 6, measuring bandwidth and latency at 8 vCPU and 32 GB. It also publishes a catalog pricing Lightsail instances, such as 2 CPU and 4 GB at $0.027 an hour, and benchmarks the AWS global network for bandwidth and latency between every region.
Sources & verification
| Source | What was checked | Last checked |
|---|---|---|
| Amazon Official | Official product page | July 10, 2026 |
| Amazon About Aws Global Infrastructure | About Aws Global Infrastructure | July 10, 2026 |
| Amazon Apprunner Pricing | Pricing and plans | July 10, 2026 |
| Amazon Cloudfront Features | Cloudfront Features | July 10, 2026 |
| Amazon Compliance Programs | Compliance Programs | July 10, 2026 |
| Amazon Compute Sla | Uptime SLA | July 10, 2026 |
| Amazon Developer docs | Dg Manage Autoscaling | July 10, 2026 |
Every fact on this Amazon Web Services (AWS) page is tied to a named source and a verification date. Freshness-sensitive figures trace to the sources above; verify against the vendor before relying on them.
Explore Amazon Web Services (AWS)
Every page on Amazon Web Services (AWS) in one place, you are on performance.