Measured results

AWS: ALB → 2 × m5.large → db.r5.large MySQL in us-east-2, measured at 10–1,000 RPS. GCP: external HTTP(S) load balancer → 2 × n2-standard-2 → Cloud SQL MySQL 8.0 in us-central1, measured at 300–500 RPS, with 5 of 6 P95 holdout rungs within 20% and measured P95 slower than predicted at every load. Details: /accuracy#gcp-holdout

Multi-Cloud Burst Simulation Report

AWS vs GCP vs Azure vs OCI vs DigitalOcean

An illustrative burst-traffic scenario across five cloud providers. No cloud infrastructure was provisioned for this report. The separate AWS measurements linked above do not validate this five-provider scenario.

Scenario configuration

Each configuration contains three compute instances, one database and one load balancer.

Provider / regionComputeDatabaseLoad balancer
AWS / us-east-13 × m5.large — 2 vCPU / 8 GB eachRDS db.r5.large, MySQLApplication Load Balancer
GCP / us-central13 × e2-standard-2 — 2 vCPU / 8 GB eachCloud SQL db-n1-standard-2Cloud Load Balancing
Azure / East US3 × Standard_D2s_v3 — 2 vCPU / 8 GiB eachSQL Database Standard S2Standard Load Balancer
OCI / us-ashburn-13 × VM.Standard.E4.Flex, configured at 2 OCPU eachMySQL HeatWave DB SystemOCI Load Balancer
DigitalOcean / nyc13 × s-4vcpu-8gb — 4 vCPU / 8 GB eachManaged DB db-s-2vcpu-4gbStandard Load Balancer

AWS m5.large and GCP e2-standard-2 both provide 2 vCPU and 8 GB RAM per instance. Equal nominal CPU and memory do not establish equal real-world performance. These configurations use different database products and model assumptions; this is not a controlled hardware performance comparison.

The engine evaluates ten simulation steps: 1,000, 1,000, 1,100, 1,200, 1,400, 1,600, 2,000, 2,000, 2,000 and 2,000 offered requests per second. Simulation steps are not measured elapsed minutes.

Cost estimates

list-price estimates (June 2026), ±10% unit-rate check

The figures below are engine-estimated hourly totals for this scenario, including modeled usage charges. They are not resource-only subtotals, measured bills or provider quotes.

ProviderBaseline: 1,000 offered RPSPeak: 2,000 offered RPS
AWS$2.33/hr$4.12/hr
GCP$2.16/hr$3.66/hr
Azure$1.97/hr$3.53/hr
OCI$0.43/hr$0.55/hr
DigitalOcean$0.400/hr$0.425/hr

The GCP figures are incomplete modeled subtotals: this fixture does not supply the Cloud SQL billing identity required by the current engine, so its database charge is excluded. The zero database entry is not a zero-cost database price. Do not use these GCP figures as a complete architecture budget.

The ±10% check concerns supported compute and database unit rates against published list prices, not the accuracy of these whole-scenario totals. It does not validate assumed utilization, usage volumes, network charges, storage, discounts or negotiated prices. Zero-price and request-billed SKUs are excluded from that check. See cloudworldmodel.ai/fidelity for the scope and exclusions.

These totals do not establish equivalent delivered performance or a best-value provider ranking.

Methodology and limitations

Cloud World Model generated this illustrative scenario using a fixed-seed simulation. The engine models capacity, resource behavior and usage-based costs without running the workload on the five providers.

Cross-provider latency, throughput, error-rate and scaling-speed rankings are not reported here. GCP, Azure, OCI and DigitalOcean performance estimates have not been calibrated against owned measurements for this scenario. The separate measured AWS evidence applies only to the architectures, workloads and conditions documented at cloudworldmodel.ai/accuracy; it is not a measurement of this report's burst scenario.

Use these estimates to explore assumptions before testing your actual architecture. Validate performance with workload-specific load tests and costs with current provider pricing and your usage assumptions.

Simulate your own architecture before deploying

Drop in your own resources, run burst traffic patterns, and explore modeled AWS, GCP, Azure, OCI and DigitalOcean configurations — no cloud account required.

Try Cloud World Model → https://www.cloudworldmodel.ai