AWS vs. Azure vs. Google Cloud: Which free tier is best?

Every major cloud company now has a free tier, with Google recently rolling out no-cost VMs and other services. Here's how they compare

AWS vs. Azure vs. Google Cloud: Which free tier is best?

Who doesn't like free stuff? Google knows we all do.

This week Google unveiled a new free-to-use tier for Google Cloud Platformaimed at users with modest compute, storage, database, and networking demands.

In the past, the company has offered a limited-time set of free usage credits for new Google Cloud Platform customers. That offer hasn't gone away; new sign-ups can still get $300 credit good for the first 12 months (previously 60 days).

What's totally new is Always Free, a no-cost usage tier available for many Cloud Platform products. It's ideal for a scrappy startup or indie developer spinning up prototypes or launching private betas, or even for minimal public applications like low-bandwidth static sites.

Here's what you need to know about the free end of Google's Cloud Platform pool -- and how the reigning cloud competition stacks up in key areas.

What you get from Google's cloud

Always Free doesn't only provide compute and storage. The total list of products with free tiers spans pretty much everything you'd need to create modern cloud-based software, including Container Engine and Cloud Functions.

Naturally, all of them have fairly low usage ceilings. With Compute Engine, for instance, you're limited to a single f1-micro instance: 20 percent of one virtual CPU (albeit burstable above 20%), 600MB of memory, and available only in the United States (sorry, Europe!), with 30GB of persistent storage per month (5GB of snapshot storage) and 1GB of network egress per month from the U.S. to each discrete egress destination (U.S., Asia, and Europe). For static sites or minimal applications, that's a fine fit, but anything more ambitious will feel like it's in the glove compartment.

Don't expect to build truly cutting-edge stuff on the free tier alone. With Container Engine, which runs Kubernetes, you have only a basic five-node cluster -- and while the clustering is free, each node in the cluster costs standard Compute Engine costs.

A few other fancy goodies are included for free, like the Cloud Vision, Speech, and Natural Language APIs. Those are also usage-limited -- Speech maxes out at 60 minutes per month, for instance. At this tier, they're mainly for experimentation, not production use.

The recently announced Container Builder is positioned for the long term as a general-purpose software assembly pipeline, as it can do more than whip up Docker containers. If your needs are modest, you might be able to get away with staying under the limit of 120 minutes of build time per day.

How Amazon AWS compares

Amazon has both an always-free tier and a first-12-months-free offer, but its always-free tier is missing some features found in Google's. It has no virtual machines, for instance; you only get those free with Amazon's 12-month offer. But the machines you get in the limited-time deal are slightly beefier, with 1GB of RAM, and they allow Windows Server instances, even if in minimal incarnations. With Google, Windows Server images are cost-plus, period.

The AWS always-free tier also doesn't include bandwidth -- that's only available in the 12-month offer, which also includes 15GB of bandwidth out, aggregated across all used services. You do, however, get up to 1 million free AWS Lambda requests per month, plus 4,000 Step Functions, which helps stage simple applications on a budget.

A few other other big developer features are available on the AWS free tier, including AWS CodeBuild, CodeCommit, and CodePipeline. All are time- and resource-limited, though; CodeBuild only gives you 100 build minutes per month.

But AWS's always-free tier doesn't include access to any of its machine learning or AI APIs like Amazon Rekognition, Polly, or Lex for visual, object detection, text-to-speech, and conversational interfaces, respectively. That said, they have relatively low pricing to start with: Rekognition, for instance, begins at $1 per 1,000 images. (Rekognition and Polly do have some free offerings, but they're only available in the first twelve months after sign-up.)

How Microsoft Azure compares

Azure also has free credit for new customers and a free service tier. But as with Amazon, compute and bandwidth aren't available as free-tier options. Instead, you get a plethora of smaller items, many of them development-oriented giveaways for users of higher tiers.

It's still possible to do scaled-down app development and deployment on the free tier. App Service provides up to 10 apps per account, and you can have up to five Visual Studio Team Services users with basic access to CI and version control at no cost. You also get virtual networks at no added cost, although you pay for the traffic itself. If you end up needing Azure Active Directory, the free tier provides up to 500,000 directory objects gratis.

When it comes to machine learning features, Azure has some production-constrained free tier options in Azure Machine Learning Studio. You get only 10GB of storage, a maximum of 100 modules per experiment, and one hour execution time per experiment. Also, there's no SLA on the free tier for Studio and, thus, no guarantees anything you run on it will complete in a timely fashion.

[This article was amended to add some details about Google's offerings, and about Amazon's AI products.]