Google Cloud Platform (GCP) vs Amazon Web Services (AWS) vs Microsoft Azure - Cloud IaaS - Price Comparison

I’ve decided to share some public cloud Infrastructure as a Service (IaaS) compute instance cost analysis that I recently created as part of a project for one of my clients.  When choosing an IaaS provider there are obviously many things to consider beyond just compute instance pricing.  Other factors such as storage, network bandwidth, snapshot and replica options and many other factors (and costs) come into play.  Each of these providers offers many different services that may be of differing value to potential customers.

Conclusions (up front for you TL/DR folks):

• The commonly accepted wisdom is that these providers are locked in a price war and that they have all closely matched each others pricing.  Nothing could be further from the truth.  Instances from Microsoft Azure are dramatically more expensive that Amazon Web Services and Google Cloud Platform no matter how you slice the data.  Google Cloud Platform and Amazon Web Service pricing looks close if you compare total three year costs.  However, how you get those numbers to be close (write AWS big checks upfront) is dramatically different.
• Based on the numbers we chose for cost of capital (5%) and likely future IaaS price cuts (15% /yr), AWS does in many cases offer the lowest cost three year option IF you are willing to pay substantial amounts upfront.
• Google Cloud Platform offers extremely competitive pricing with no upfront purchase needed.
• Windows is expensive.  In some cases the cost difference in a Linux instance and a Windows instance exceeds the cost of the Linux instance itself.  Think about that for a moment.  The cost of your OS choice can more than double the cost of your instance.  I love Microsoft.  I love Windows.  I hope this changes.

Update – 12/6/2016 – A Microsoft rep posted this comment on my LinkedIn post of this article.  Keep this in mind as you compare prices.

If potential Azure customers talk to their local Microsoft sales rep they can chose to buy via a so called “Compute Pre Purchase” option. It will give you up to >45% savings for modern compute instances depending on the location and instance family. You need to decide for a location, instance type and pay for one year upfront but still might be appropriate for many use cases. Microsoft will very soon offer an easier way to leverage those savings and offer more options as well as longer term periods, etc. very soon.

Methodology:

In order to simplify some of the discussion for this purpose of this post, we’ve made the following assumptions.

• We will look at only four similar instance sizes.
• We will not consider storage, bandwidth or other costs.  Perhaps that will be a discussion for another post.
• We will look at the cost difference between running Linux and Windows instances.
• We will consider and attempt to model the different purchase options available from each provider.
• We will compare the costs for running these compute instances for both one and three year terms.
• We will assume 100% sustained use during the entire period considered.

Instance sizes:

• Small – At least 1 CPU core / ~4GB of RAM
  • Specific instances we chose to compare: AWS: t2.medium / GCP: n1-standard-1 / Azure: GP A2
• Medium – At least 2 CPU cores / ~8GB of RAM
  • Specific instances we chose to compare: m4.large / GCP: n1-standard-2 / Azure: GP A5
• Large – At least 4 CPU cores / ~16GB of RAM
  • Specific instances we chose to compare: AWS: m4.xlarge / GCP: n1-standard-4 / Azure: GP A6
• Extra Large – At least 8 CPU cores / ~30GB of RAM
  • Specific instances we chose to compare: AWS: m4.2xlarge / GCP: n1-standard-8 / Azure: GP A7

There is no perfect way to compare things that are not identical.  So, we have chosen what we believe to be fairly similar instance types to compare.

Provider Pricing Model Discounts:

Each provider offers ways to purchase instances in order to save some money.

Amazon Web Services: AWS offers a variety of purchase options.  These options can result in significant savings.  Explaining how reserved instances work is beyond the scope of this article.  For more detail on this topic go here: https://aws.amazon.com/ec2/purchasing-options/reserved-instances/ .  In general, the longer term you are willing to commit to and the more you are willing to pay upfront, the higher the discount you can get.

Azure: Microsoft Azure offers a flat 5% discount if you are willing to pre-pay for 12 months of service upfront.  https://azure.microsoft.com/en-us/offers/ms-azr-0026p/.  The 5% Microsoft discount is frankly not very enticing compared to the significant discounts you can get from AWS for prepayment, and compared to the discounts you get from Google for simply using instances on a sustained basis.  Since a three year upfront purchase is not possible, when we modeled Azure three year costs we did so by estimating the cost of three annual purchases.

Google Cloud Platform: Google offers great discounts for sustained use.  You don’t need to pre-purchase anything, you get the discounts automatically.  The discounts are very substantial.

Cost of capital:

For purposes of this post we also wanted to consider the cost of capital.  It is also not reasonable to compare spending a large sum of money upfront with spending no money upfront and simply paying for what you use on an ongoing basis.  So, for purposes of this discussion we are going to assign a relatively arbitrary 5% annual cost of capital to options where prepayments are considered.

Expected future IaaS price reductions:

The costs of public cloud IaaS continue to drop.  For purposes of these calculations, when we look at one year costs we will assume that no price drops will happen during the middle of our one year term.  For purposes of our three year estimates, we will assume that a 15% price reduction will happen at the end of year one, and another 15% price reduction will happen at the end of year two.   Obviously, these are best guess estimates and we could easily be wrong.

Shameless Plug:

If your business needs help figuring out how to best architect public cloud infrastructure, we would love to help.

Raw data:

Linux – 1 Year

Note: A 5% cost of capital has been used for these calculations where an upfront purchase was required.

Linux – 3 Years

Note: A 5% cost of capital has been used for these calculations where an upfront purchase was required.  A 15% annual cost reduction has been estimated.

Windows – 1 Year

Note: A 5% cost of capital has been used for these calculations where an upfront purchase was required.

Windows – 3 Years

Note: A 5% cost of capital has been used for these calculations where an upfront purchase was required.  A 15% annual cost reduction has been estimated.