How vmprices collects its prices

Every price on this site comes from the providers' own published price data. Nothing is estimated, scraped from third parties or adjusted.

What a price is

• List prices in US dollars for Linux virtual machines, as each provider publishes them. Your invoice can differ: taxes, negotiated discounts, credits, free tiers and Google's sustained-use discounts are not included.
• A month is 730 hours. The monthly price is the hourly price × 730, plus a 10 GB boot disk when "Include a 10 GB boot disk" is on (it is by default). Per vCPU and per GB divide that monthly price by the machine type's vCPUs or memory.
• Spot prices are what each provider lists for spot (preemptible) capacity in a region. Spot VMs can be stopped by the provider at any time, and the price can change: Google says GCP Spot prices can change up to once every day.
• 1-year prices are Google Cloud's 1-year resource-based committed use prices and Azure's 1-year reserved instance prices, per hour. AWS 1-year prices are not collected yet.

The matrix: sizes down, zones across

The matrix on the front page works like a low-fare calendar. Each row is a size, each column a zone, and each cell the cheapest offer for that size in that zone, from the clouds and pricing you pick. It shows the machine type and cloud that won, and a click shows the next cheapest.

Sizes

A row is a number of vCPUs and an amount of memory. A machine type is in a row when it has at least the row's vCPUs and memory, fewer vCPUs than the next level and less than twice the row's memory. So every machine type is in at most one row, and a slightly bigger one still counts: a 2 vCPU type with 7.5 GB is in "2 vCPU · 4 GB", a 24 vCPU type with 96 GB in "16 vCPU · 64 GB". Each row's second line says which vCPU and memory figures its cells actually hold. The rows run from 1 vCPU to 192 vCPU, with 1, 2, 4 and 8 GB per vCPU (and the micro sizes, down to 0.5 GB, at 1 and 2 vCPU). GPU and accelerator types are left out, and so are types with 16 GB per vCPU or more, or under 1 GB per vCPU above 2 vCPU; they are on the machine type view and pages.

Architecture, CPU vendor, shared or dedicated cores and a highest monthly price are filters on top of this. "x86 and Arm, separate rows" gives every size an x86 row and an Arm row.

Zones

Columns are zones: the same geography on every cloud. The providers name their regions differently (Google's europe-west3 is Frankfurt and europe-west8 is Milan, AWS's eu-central-1 is Frankfurt too), so a zone is built from where each region is documented to be, never from its name: Google lists a city for each region, Microsoft a "physical location" (a city, a state or a country), AWS puts the place in the region's name. In the United States, where AWS and Azure name only the state, the states decide: Virginia, South Carolina and Ohio are us-east; Iowa, Illinois, Texas and Wyoming us-central; the Pacific and mountain states us-west. In Europe, Ireland, the United Kingdom, Paris and the Benelux are eu-west; the Nordic countries eu-north; Germany, Switzerland, Austria and Poland eu-central; Italy, Spain and Marseille eu-south. Where a provider names only a country (AWS "Canada (Central)", Microsoft "Netherlands"), the country decides. The coarser level groups the zones by part of the world.

Sources: Google Cloud: Regions and zones (location of each zone), AWS: Regions (the region names), Microsoft: Azure regions list (physical location), and AWS's launch post for Aragón.

Sources

Google Cloud

The Cloud Billing Catalog API, Compute Engine service 6F81-5844-456A, about 33,500 SKUs. A machine type costs its vCPUs × the family's per-vCPU price plus its memory × the per-GB price in that region; spot SKUs are named "Spot Preemptible …". Each SKU's serviceRegions says which regions it applies to. Shared-core E2 types bill part of a vCPU (e2-micro a quarter, e2-small half, e2-medium one). Machine shapes and the regions each type is offered in come from the Compute Engine API (gcloud compute machine-types list). Not included yet: GPU (accelerator-optimized) types, types with bundled local SSD (-lssd, Z3), X4, custom machine types and sole-tenant nodes.

Boot disk: a 10 GB Balanced Persistent Disk, or for the families that only boot from Hyperdisk (N4, N4D, N4A, C4, C4D, C4A, C4N, H4D, M4, M4N) a 10 GB Hyperdisk Balanced disk with its default 3,060 IOPS and 155 MiB/s (3,000 IOPS and 140 MiB/s are free).

History: the Catalog API also returns past prices for a time range. It accepted ranges back to 1 October 2026, so GCP history starts there.

AWS

The public files behind AWS's own EC2 pricing pages: on-demand Linux prices per region (b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/…/ec2-ondemand-without-sec-sel/…/Linux, shared tenancy, no pre-installed software), the Linux column of the Spot pricing page (website.spot.ec2.aws.a2z.com/spot.json) and EBS volume prices. Processors and architecture come from the public bulk Price List. Boot disk: 10 GB gp3. AWS's own spot price history (90 days) needs an AWS account and is not used.

Azure

The Azure Retail Prices API: Linux consumption and Spot meters of Virtual Machines, 1-year reservations, and managed disk prices. The API has no vCPU or memory figures, so those come from Microsoft's own pricing calculator data; sizes the calculator does not list yet are left out. CPU vendor and architecture follow Microsoft's size naming convention ("a" AMD, "p" Arm). Boot disk: Azure bills managed disks by size tier, and the smallest Standard SSD that holds 10 GB is E3 (16 GiB).

How often

A scheduled job checks every provider once a day, from 04:00 UTC. A price is stored again only when it changes, so the history of each cell is its first price and every change after it. Machine type and region pages show when each price last changed.

History

The history views are built once a day after the check, from those stored changes. For every size row, region, architecture and core type they keep the cheapest VM price on each day it changed, and which machine type it was; a price holds until the next change, so lines are steps. A line for a zone or part of the world is the cheapest (or, where it says so, the median) of its regions on each day. The history leaves out the boot disk, whose price has no stored history, so its prices are a little under the matrix's default. Each cloud's history starts with its first stored price: Google Cloud's catalog returned prices back to 1 October 2026, AWS and Azure start with the first check on 24 October 2026. Biggest movers compares each price with its price at the start of the window, or with its first stored price where history starts inside the window.

The latest check (2026-10-25)

Data

The pages read JSON files you can use too: /api/matrix/<gcp|aws|azure>/<spot|ondemand|commit1y>.json (hourly prices per machine type and region), /api/history/<provider>/<machine type>.json, /api/series/<provider>/<model>.json (the cheapest price of each size and region over time) and /api/movers.json.