This is what Snap is paying Google $2 billion for

Snap, which is set to IPO this week, signed a $2 billion, five-year contract with Google for its cloud services, which makes Snap Google’s largest customer of its cloud platform.

For context, Google’s cloud business was previously estimated to have an annual runrate revenue of about $1 billion, according to RBC Capital Markets analyst Mark Mahaney, trailing Amazon Web Services’ $12.2 billion and Microsoft Azure’s estimated $2.7 billion. (Snap also has a $1 billion contract with AWS.)

But what is Snap actually getting from Google for all that money?

Google wouldn’t say exactly which services Snap has signed up for, but there are four key products that fall under Google’s cloud services: Cloud Storage for storing data on servers managed by Google; Compute Engine for retrieving and managing data; App Engine for developing and running applications; BigQuery for data analysis; and a suite of machine learning tools.

We know Snapchat was built on App Engine, which basically allows clients to host their main software on datacenters managed by Google. That differs from just hosting your own servers, since cloud services allow a company to make use of more servers as needed, allowing the app to run faster and more efficiently.

App Engine is one of Google’s core cloud services. It provides users with tools and services to build software, ensure its security and test new features. It also allows apps to handle an increasing amount of traffic. Other companies that use App Engine include Best Buy and enterprise cloud phone system Dialpad.

Here’s how different tools may be useful to Snap based on how other companies use Google’s cloud offerings.

For example, Google cloud customer Spotify uses BigQuery to analyze user listening data, according to a report from analyst firm Enterprise Strategy Group. ESG cited a query for top tracks by age, gender and market: What took five hours with other tools, Spotify’s use of BigQuery took about eight minutes.

Then there’s Dialpad, a startup that sells workplace communications services like UberConference, which runs on Google’s cloud service, according to CEO Craig Walker. That includes using App Engine, which the company uses for such essential processes as initiating and routing users’ video, voice and conference calls.

But Dialpad processes the actual voice and video calls through its own data centers so it can more closely control the quality of calls, according to Walker, a former Googler. Even a tiny lag in a voice call can upset users, so controlling the servers directly is important, he said.

That means companies that make heavy use of Google cloud won’t necessarily use it for everything, since there are some limitations to using the service. Dialpad’s approach when it comes to calls could be relevant to Snapchat, which lets users place calls from within the app.

Snap might also make use of Google’s machine learning software, which largely refers to more sophisticated programs that can automate certain tasks.

Snap uses a machine learning system right now that helps it to improve ad targeting. It predicts which ads users are more likely to respond to and shows the ad to the right users. The goal is to increase a certain action, like downloading an app, instead of just showing more impressions.

However Snap uses Google cloud, the company is clear that the product is critical to the startup’s operations. The company’s S-1 filing states, “Any disruption of or interference with our use of the Google Cloud operation would negatively affect our operations and seriously harm our business.”

This article originally appeared on Recode.net.