How CloudCostMatrix Calculates Pricing
Every number on this site is directional — useful for architecture planning and provider selection, not a substitute for a quote from a provider's own calculator. This page explains exactly how the engine works, what it approximates, and how fresh each provider's catalog is, so you can judge how much to trust any given figure.
How the calculation works
- Nearest-shape matching. When you request e.g. 4 vCPU / 16 GB RAM, the engine picks the closest published instance shape from each provider's catalog — it does not interpolate between shapes, so small requests can land on a larger-than-requested instance if that's the nearest match.
- 730 hours/month. All hourly rates are converted to monthly using a flat 730 hours (the average length of a month), not the exact days in the current calendar month.
- Regional multipliers. Prices are benchmarked in each provider's primary US region, then adjusted by a flat regional multiplier (e.g. Europe ×1.10, Asia Pacific ×1.16) rather than fetching real region-specific rates for every region.
- Reserved & Spot pricing is derived, not fetched, even for live-synced providers. Every public pricing API used here (Azure, Oracle, Linode) returns on-demand rates only — 1-year, 3-year, and Spot/preemptible rates are computed as fixed multipliers of the live on-demand rate (e.g. 3-yr ≈ ×0.40–0.67, varying by provider). AWS's reserved/spot rates come from the seeded benchmark. This is disclosed per-provider in the table below.
- Coverage gaps are excluded from ranking, never faked. When a provider genuinely doesn't sell a requested combination (e.g. managed SQL Server on DigitalOcean, an Archive storage tier on Linode), the comparison table shows "Not offered" and that provider is excluded from the cheapest-provider ranking for that configuration — it is never assigned a $0 or estimated substitute cost.
- Currency conversion uses fixed exchange rates (7 currencies supported), not a live FX feed. Rates are updated periodically, not in real time — treat non-USD figures as approximate.
Per-provider pricing freshness
Live — fetched from the provider's public pricing API on an automated schedule. Verified — manually reconciled against the provider's published pricing page on the date shown. Estimate — a relative-positioning estimate not yet reconciled against real published rates; treat these numbers as the least reliable on the site.
| Provider | Status | Source | Notes |
|---|---|---|---|
AWS | Live · synced Oct 29 | https://aws.amazon.com/s3/pricing/ | S3 capacity tiers are fetched live. S3 request/operation rates, EC2 compute, and internet egress use the seeded 2026 benchmark — the AWS Price List bulk EC2 index is ~480MB, impractical to sync on a schedule. |
Azure | Live · synced Oct 29 | https://prices.azure.com/api/retail/prices | Compute, internet egress, and Blob Storage capacity and operation rates are fetched live from the Azure Retail Prices API (eastus). Storage is pinned to Locally-Redundant (LRS) General Block Blob v2 meters and to the marginal first-volume tier — geo-redundant (GRS/ZRS) redundancy is not modelled. Reserved-instance (1-yr/3-yr) and Spot rates are derived as fixed multipliers of the live on-demand rate, not fetched directly — Azure does not publish those via this API. |
GCP | Live · synced Oct 29 | https://cloudbilling.googleapis.com/v1/services/6F81-5844-456A/skus | Compute, object-storage capacity, and internet egress are fetched live from the Cloud Billing Catalog API (us-east1). GCP bills compute per vCPU-hour and per GB-hour per machine family, so predefined machine types are priced by multiplying those live family rates by a static shape table — the six shapes match the benchmark catalog exactly. Storage is pinned to regional (South Carolina) capacity; dual-region and multi-region rates are not modelled. Internet egress uses the Americas group rate, whose three published tiers are flattened onto this catalog's two-rung ladder. Object-storage operation rates, load balancer, static IP, and Kubernetes figures still carry the seeded 2026 benchmark — Cloud Storage SKUs publish no billing unit, so their operation rates cannot be converted honestly. Reserved and Spot rates are derived as fixed multipliers of the live on-demand rate. |
Oracle | Live · synced Oct 29 | https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/ | Compute, storage, egress, and load balancer pricing sync live from OCI's public cetools price list (converted from OCI's per-OCPU billing unit — 1 OCPU = 2 vCPU on the x86 shapes modeled here). Managed database pricing is not synced live: OCI prices MySQL/Autonomous DB per ECPU, a different unit from this app's per-instance model, so database figures use the seeded 2026 benchmark. Reserved (1-yr/3-yr) and Spot rates are derived as fixed multipliers of the live on-demand rate — OCI's price list does not publish those directly. |
IBM | Live · synced Oct 29 | https://globalcatalog.cloud.ibm.com/api/v1/is.instance | Compute pricing syncs live from IBM Cloud's Global Catalog. IBM publishes no per-profile VPC list price — the profile entries carry no pricing and the Gen3 plan base metric is zero — so each profile rate is composed from the live vCPU-hour and GB-hour component rates IBM actually bills VPC servers against (the Standard Gen 2 / advanced-vsi plan): vCpu × vCPU-hour + GB × GB-hour. The composition is ours even though the rates are IBM's. The shape names were corrected to IBM's real profiles: the benchmark labelled the 2 vCPU/4 GB and 32 vCPU/128 GB points "bx2-2x4" and "mx2-32x128", neither of which exists (bx2 is 1:4 and mx2 is 1:8); they are now cx2-2x4 and bx2-32x128. Reserved (1-yr/3-yr) rates are derived as fixed multipliers of the live on-demand rate; there is no spot tier (gated off in PROVIDER_CAPABILITIES). The Windows licence surcharge, managed databases, object storage, Kubernetes, load balancer, and static IP pricing all carry the seeded 2026 benchmark. |
DigitalOcean | Live · synced Oct 29 | https://www.digitalocean.com/pricing/droplets | Compute pricing syncs live from DigitalOcean's public Droplet pricing page for the bundled-plan families (Basic, CPU-Optimized, General Purpose, Memory-Optimized) across a 2–32 vCPU envelope. DigitalOcean's /v2/sizes API is deliberately NOT used: it returns a filtered catalog whose largest non-GPU Droplet is 4 vCPU, which would match a 32 vCPU request to a 4 vCPU shape and render DigitalOcean roughly 8x too cheap. Rates are derived from each plan's flat monthly cap (monthly / 730), not the published $/hr meter — DigitalOcean prices the hourly meter above the monthly-equivalent rate so a full month billed hourly never undercuts the bundled price. v5 Droplets are excluded: they bill per configured resource hourly with no monthly cap, so a monthly-equivalent rate is not a like-for-like comparison. Spaces object storage and Managed Databases are not published on this page, so those sections carry the seeded 2026 benchmark. Spaces has a single storage class (no Cool/Cold/Archive tiering), DigitalOcean offers no SQL Server, and it does not sell Windows Server licensing on Droplets. Bundled outbound transfer genuinely varies by plan (500–10,000 GB). This catalog models one flat per-instance allowance taken from the reference plan, not a per-shape value. |
Alibaba | Estimate — not yet list-verified | https://www.alibabacloud.com/pricing | Pricing is a relative-positioning estimate anchored to AWS list price. Alibaba Cloud's pricing API requires signed (HMAC) requests — not yet wired up. |
Linode | Live · synced Oct 29 | https://api.linode.com/v4/linode/types | Compute, managed database, object storage, NodeBalancer, and Kubernetes control-plane pricing sync live from Linode's public v4 API. Two figures are not exposed by that API and are carried from the seeded benchmark: managed-database storage cost per GB (disk is bundled into the instance price) and the egress overage rate. The per-Droplet-equivalent pooled bandwidth allowance genuinely varies by instance size (1,000–20,000 GB) — this catalog uses one representative mid-size value rather than a per-shape allowance. |
OVHcloud | FX-converted · Oct 29 | https://api.ovh.com/1.0/order/catalog/public/cloud?ovhSubsidiary=IE | Compute and object-storage rates sync live from OVHcloud's public cloud catalog in EUR (IE subsidiary). OVHcloud publishes no USD subsidiary, so every figure is converted to USD at the ECB reference rate recorded at sync time — the exact rate and date are shown with the freshness label, and this provider is labelled FX-converted, never plain live. Reserved (1-yr/3-yr) rates are derived as fixed multipliers of the live on-demand rate; OVHcloud publishes no commitment rates for these flavors via this catalog, and there is no spot tier (gated off in PROVIDER_CAPABILITIES). Object-storage operation rates are not published in a convertible billing unit and carry the seed benchmark, as does the per-instance Windows licence surcharge (OVHcloud prices Windows Server per vCore-hour on a separate SKU). Managed databases, load balancers, and static IPs are not synced and carry the seeded 2026 benchmark. Outbound internet bandwidth is unlimited and free on every Public Cloud plan. Converted at 1 EUR = 1.113 USD (ECB reference rate dated 2026-10-29, via https://api.frankfurter.app/latest?from=EUR&to=USD). |
Vultr | Live · synced Oct 29 | https://api.vultr.com/v2/plans | Compute pricing syncs live from Vultr's public v2 plans API for the Cloud Compute (Regular/High Frequency/High Performance) and Optimized Cloud Compute families, across a 2–32 vCPU envelope. Vultr bills hourly with a 672-hour monthly cap, so each shape's steady-state rate is derived from its listed monthly price (monthly / 730) rather than the capped hourly meter — using the raw hourly rate would understate a full month. Managed-database pricing syncs live from the authenticated v2 databases/plans API (VULTR_API_KEY) across a 1–8 vCPU envelope. Vultr prices PostgreSQL and MySQL identically per shape, so both are mapped from the same rate; database storage cost per GB and the multi-AZ (replica) multiplier are not exposed by that endpoint and carry the seeded benchmark. If VULTR_API_KEY is not configured, the database section also falls back to the seed. Vultr sells no reserved or spot instances, so both commitment rows clone the flat on-demand rate (gated off in PROVIDER_CAPABILITIES). Object storage, networking, and Kubernetes carry the seeded benchmark — object storage at $18/TB-month with a 1 TB minimum and a single storage class, a representative 2 TB per-instance transfer bundle (plans range 0.5–15 TB), and approximated Windows/load-balancer/static-IP rates. |
Affiliate links & ranking independence
CloudCostMatrix carries affiliate links to some providers — see the full disclosure for which ones and what commission structure applies. The cost calculations, rankings, and "cheapest provider" verdicts on this site are computed entirely by the pricing engine described above, from the published list-price catalogs — no affiliate relationship touches that math. If a provider we don't have a commercial relationship with is cheaper for your workload, that is what the calculator will show.
Supported regions
5 region multipliers: US East, US West, Europe, Asia Pacific, South America. Regional multipliers are applied uniformly across all providers and are not provider-specific.