Cloud decisions / Interactive field guide
Which cloud fits the workload—not just the headline price?
Normalize the workload and quote boundary first. Then inspect the modeled line items without pretending that cost alone selects an architecture.
Current example
AWS and Azure and Google Cloud tie under the entered quotes
Entered quotes, not live prices or equivalent-instance performance. Compute plus billed outbound transfer only; excludes storage, requests, support, tax, commitments, credits, cross-zone and managed-service fees. Effective transfer quotes must already reflect units, tiers and allowances. Residency, identity, availability, operating skills and performance still constrain a provider choice.
- Highest minus lowest modeled total
- $0.00 USD
Initial quotes are deliberately equal: 730 instance-hours, one replica, 1,000 normalized billed GB outbound, $1 per instance-hour and $0.10 per outbound GB for every provider. These are not current prices.
Normalize the workload
- Hours in the comparison window
- 730
- Equivalent instance count
- 1
- Normalized outbound billed GB
- 1000
AWS quote
Azure quote
Google Cloud quote
- Google Cloud quote / instance-hour (USD)
- 1
- Google Cloud effective outbound quote (USD/GB)
- 0.1
- Google Cloud modeled subtotal
- $830.00 USD
02 / Follow the flow
Four boundaries. One connected explanation.
Each card explains a node in the diagram. Highlights show which boundaries contribute to the current result.
Workload
Compare an outcome before comparing an instance
Specify throughput, latency, recovery, data residency and numerical quality where relevant. Equal instance counts or accelerator labels do not establish equal delivered work. Confirm region availability and quotas before a pricing exercise becomes a deployment assumption. A workload-normalized pilot can show whether the candidate configurations meet the same acceptance criteria; this arithmetic reader cannot.
Compute
Keep the quote and responsibility boundary explicit
An instance-hour quote may exclude managed control planes, operating-system licensing, support, commitments or attached services. Managed container and AI offerings also divide operational responsibilities differently. The reader multiplies an entered effective compute quote by hours and replicas; it does not fetch prices, resolve instance families or assign a value to the team’s operating skills.
Transfer
Follow where bytes actually cross a billable boundary
Internet, cross-zone, cross-region and inter-service transfer may be priced differently. Providers can use different units, tiers and allowances. Normalize the physical workload and its billable classes, then enter an effective rate for the outbound scope shown here. Moving data to a managed service can shift the bill rather than remove it, so keep excluded network and storage costs visible in a real decision record.
Decision
Price is one constraint among several
A lower modeled subtotal cannot override an unavailable region, an unmet recovery target, an unacceptable identity boundary or an unsupported workload. Evaluate residency, auditability, managed-service fit, team competence and exit costs. Terraform can make intent reviewable but does not erase provider-specific IAM, networking, storage and operating behavior. Record the evidence and the conditions under which the choice should be revisited.
# Same scope for every entered provider quote
compute_subtotal = hours * replicas * quoted_compute_USD_per_instance_hour
egress_subtotal = billed_egress_GB * quoted_egress_USD_per_GB
modeled_total = compute_subtotal + egress_subtotal
spread = max(aws_total, azure_total, gcp_total) - min(aws_total, azure_total, gcp_total)
# Compare operational constraints before selecting a provider.03 / Keep the model honest
Model assumptions
Compute subtotal = hours × equivalent instances × entered instance-hour quote
Outbound subtotal = normalized billed GB × entered effective outbound quote
Modeled provider subtotal = compute subtotal + outbound subtotal
Spread = largest modeled subtotal − smallest modeled subtotal
Equal placeholder defaults intentionally produce no price winner.- All defaults are equal placeholders. This is not a live pricing service, procurement quote, total-cost-of-ownership model or recommendation to choose a provider.
- Scope is compute plus one normalized outbound transfer class only. Storage, request charges, support, taxes, credits, commitments, managed-service fees and other transfer classes are excluded.
- Use effective quotes that already reflect the relevant billing units, free allowances and tiers. The reader applies linear arithmetic rather than a provider’s complete billing algorithm.
- The compared configurations must independently meet the same workload, region, recovery and governance requirements. Equivalent replica counts alone do not establish equivalent performance.
- No cloud account is queried and no resources are provisioned. The comparison remains readable without JavaScript.
04 / Think it through
Questions behind the example.
Which units, regions, workload characteristics and commercial terms must be normalized before comparing real quotes?
When can outbound transfer architecture matter more than a small instance discount?
Why could a zero effective quote mean an allowance, a temporary credit or missing data rather than a suitable provider?
Primary documentation